Simulation tool and agent interaction method and system based on convolution operation

Abstract

The invention provides a convolution operation-based simulation tool and agent interaction method and system, and the method comprises the steps: carrying out the data analysis and data unification of a plurality of time series data based on the time series data generated by a simulation tool, and obtaining a to-be-input data set; performing feature extraction on the to-be-input data set through a preset convolutional neural network model to obtain a first demand feature value and generate a plurality of training strategies; based on a plurality of training strategies, matching corresponding simulation tool requirements, and generating a second requirement feature value; and sending the second demand characteristic value to the simulation tool so as to complete one-time interaction between the simulation tool and the intelligent agent. The problem that the interaction efficiency of multiple agents and multiple simulation tools is low due to the fact that the prior art is limited to one-to-one and cannot process differentiated power grid data is solved. According to the method, differentiated data can be processed and analyzed, the method is suitable for different types of novel power system configurations, and the interaction efficiency between multiple agents and multiple simulation tools is improved.

Description

Technical Field

[0001] The present invention relates to the field of power grid interaction design, and in particular to a method and system for interaction between a simulation tool and an intelligent agent based on convolution operations. Background Art

[0002] The new power system has already shown the "double high" characteristics of a high proportion of renewable energy and a high proportion of power electronic equipment. This change has led to a major change in the power grid operating environment, that is, the interconnection between different power grids will become complex and diverse, causing the new power system to present high-dimensional nonlinear characteristics.

[0003] At present, the interaction between traditional power grid simulation tools and intelligent agents is a one-to-one interaction between each intelligent agent and a simulation tool. The traditional method is only effective in dealing with simple or standard power grid operation and control problems. Faced with high-dimensional nonlinear power grid architecture, traditional interaction methods cannot handle differentiated power grid data or require manual design and adjustment of data formats for each tool or data type during the interaction process. Overall, traditional interaction methods make the interaction efficiency between power grid simulation tools and intelligent agents low. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a simulation tool and intelligent agent interaction method and system based on convolution operation, which allows the intelligent agent and the power grid simulation tool to perform information sharing and strategy coordination while unifying the data format, thereby effectively improving the interaction efficiency between the intelligent agent and the simulation tool.

[0005] To achieve the above object, an embodiment of the present invention provides a method for interaction between a simulation tool and an intelligent agent based on a convolution operation, comprising:

[0006] Based on the time series data generated by the simulation tool, performing data analysis and data unification on the time series data to obtain a data set to be input;

[0007] Extracting features of the input data set by using a preset convolutional neural network model to obtain a first demand feature value;

[0008] Sending the first demand feature value to the agent to generate a number of training strategies;

[0009] Based on the plurality of training strategies, matching corresponding simulation tool requirements and generating a second requirement characteristic value;

[0010] The second requirement characteristic value is sent to the simulation tool to complete an interaction between the simulation tool and the agent.

[0011] The embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on convolution operation. By acquiring the time series data generated by the simulation tool and parsing and unifying the event continuation data and the retrograde data, the nonlinear characteristics of the data and the dynamic changes of the power system under different environments are effectively acquired. At the same time, the data format is unified to reduce the operational limitations between the data during subsequent data processing, which helps to improve the interaction efficiency between the intelligent agent and the simulation tool. Then, through the convolutional neural network model, the processed time series data is feature extracted to obtain the key features of the data to reduce the complexity of data processing, improve the data calculation efficiency and thus improve the interaction efficiency between the intelligent agent and the simulation tool. Then, the feature values ​​obtained by feature extraction are sent to the intelligent agent, so that the intelligent agent trains the corresponding training strategy, and the training strategy is converted into a data form acceptable to the simulation tool by matching the needs of the simulation tool, and then the data is sent to the simulation tool for simulation. The above method is used to ensure the completion of an efficient data interaction.

[0012] Furthermore, based on the time series data generated by the simulation tool, data analysis and data unification are performed on the time series data to obtain a data set to be input, including:

[0013] Based on a number of different simulation conditions, simulation is performed by the simulation tool to generate a number of time series data;

[0014] Performing dimension analysis on the time series data through a data analysis interface to obtain the time series data of different dimensions;

[0015] The time series data of different dimensions are normalized to obtain a data set to be input.

[0016] Further, feature extraction is performed on the input data set by using a preset convolutional neural network model to obtain a first demand feature value, including:

[0017] The parameters of the corresponding preset convolutional neural network model are set according to the simulation requirements; wherein the preset convolutional neural network model includes: a convolution layer, a pooling layer and an output layer;

[0018] Based on the convolutional layer, feature extraction is performed on the data set to be input to obtain a feature map;

[0019] Based on the pooling layer, reducing the dimension of the feature map to obtain a pooling output value;

[0020] Based on the output layer, a probability distribution calculation is performed on the pooled output value through a preset activation function to obtain a first demand feature value.

[0021] Further, based on the convolutional layer, feature extraction is performed on the input data set to obtain a feature map, including:

[0022] Inputting the data set to be input into the convolution layer of the preset convolutional neural network model;

[0023] Performing a weight matrix operation on the data set to be input through the convolution layer to obtain a convolution eigenvalue;

[0024] The convolutional feature values ​​are concatenated, flattened and filled with data images through the convolution layer to obtain a feature map.

[0025] The embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on convolution operations. According to the characteristics of the power grid and the simulation requirements, key features of time series data are extracted to effectively capture the instantaneous state and long-term trends of the power system, ensure that only valid data are calculated during the interaction process, reduce the transmission of redundant data, improve data computing power, and thus improve the interaction efficiency between the intelligent agent and the simulation tool.

[0026] Further, based on the pooling layer, the feature map is reduced in dimension to obtain a pooling output value, including:

[0027] Inputting the feature map into the pooling layer of the preset convolutional neural network model;

[0028] The feature map is subjected to maximum pooling and average pooling through the pooling layer to obtain a pooling output value.

[0029] An embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on convolution operations, which reduces the dimension of the feature map through a pooling layer, reduces the complexity of calculations and simplifies the features of simulation data, strengthens the main features, and weakens the secondary features. The overfitting risk of model training is reduced through maximum pooling and average pooling operations, thereby improving the operating stability of the model under complex data.

[0030] Further, based on the output layer, the probability distribution of the pooled output value is calculated by a preset activation function to obtain a first demand feature value, including:

[0031] Inputting the pooled output to the output layer value of the preset convolutional neural network model;

[0032] The probability distribution of the pooled output value is calculated by using a preset activation function of the preset activation function enhancement model to obtain a first demand feature value.

[0033] An embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on convolution operations, which enhances the nonlinear expression ability of the model through an activation function at the output layer, ensures that the intelligent agent can make a fast and effective response based on the extracted features when making control decisions, and improves the interaction efficiency between the intelligent agent and the simulation tool.

[0034] Furthermore, the first demand feature value is sent to the agent to generate several training strategies, including:

[0035] Sending the first demand feature value to the agent;

[0036] Based on different conditions and preset task requirements of the intelligent agent, strategy training is performed on the first requirement characteristic value to generate several training strategies that match the preset task requirements of the intelligent agent under different conditions.

[0037] Further, based on the plurality of training strategies, the corresponding simulation tool requirements are matched to generate a second requirement characteristic value, including:

[0038] Based on a preset data annotation tool, the training strategies are analyzed and annotated to generate probability distributions, actions, and expected rewards corresponding to the strategies;

[0039] Based on the simulation tool requirements, the probability distributions, actions and expected rewards corresponding to the plurality of strategies are converted into a data format analyzable by the simulation tool to obtain a second requirement characteristic value.

[0040] The embodiment of the present invention further provides a simulation tool and agent interaction system based on convolution operation, comprising:

[0041] Simulation tool input data processing module, convolution operation module, agent strategy generation module, strategy analysis module and simulation tool data recovery module;

[0042] The simulation tool input data processing module is used to perform data analysis and data unification on the multiple time series data generated by the simulation tool to obtain the data set to be input;

[0043] The convolution operation module is used to extract features of the to-be-input data set through a preset convolutional neural network model to obtain a first required feature value;

[0044] The agent strategy generation module is used to send the first demand feature value to the agent to generate a number of training strategies;

[0045] The strategy analysis module is used to match the corresponding simulation tool requirements based on the plurality of training strategies and generate a second requirement characteristic value;

[0046] The simulation tool data recovery module is used to send the second demand characteristic value to the simulation tool to complete an interaction between the simulation tool and the agent.

[0047] The embodiment of the present invention proposes a simulation tool and intelligent agent interaction system based on convolution operation, which obtains the time series data generated by the simulation tool through the simulation tool input data processing module, analyzes the event power data and the retrograde data, and unifies the data, effectively obtains the nonlinear characteristics of the data and the dynamic changes of the power system under different environments, and unifies the data format to reduce the operational limitations between the data during subsequent data processing, which helps to improve the interaction efficiency between the intelligent agent and the simulation tool; then, through the convolution operation module, a convolutional neural network model is used to extract features from the processed time series data, obtain the key features of the data to reduce the complexity of data processing, improve the data calculation efficiency and thus improve the interaction efficiency between the intelligent agent and the simulation tool; then, through the intelligent agent strategy generation module, the feature value obtained by the feature extraction is sent to the intelligent agent, so that the intelligent agent trains the corresponding training strategy, and the strategy analysis module matches the needs of the simulation tool to convert the training strategy into a data form acceptable to the simulation tool, and then the data recovery module of the simulation tool sends the data to the simulation tool for simulation, and the above method is used to ensure the completion of an efficient data interaction.

[0048] The convolution operation module is used to extract features of the to-be-input data set through a preset convolutional neural network model to obtain a first required feature value, including:

[0049] Convolutional model setup unit, convolution unit, pooling unit and output unit;

[0050] The convolution model setting unit is used to set the parameters of the corresponding preset convolution neural network model according to the simulation requirements; wherein the preset convolution neural network model includes: a convolution layer, a pooling layer and an output layer;

[0051] The convolution unit is used to extract features of the to-be-input data set based on the convolution layer to obtain a feature map;

[0052] The pooling unit is used to reduce the dimension of the feature map based on the pooling layer to obtain a pooling output value;

[0053] The output unit is used to perform probability distribution calculation on the pooled output value based on the output layer through a preset activation function to obtain a first demand feature value. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of the steps of a method for interaction between a simulation tool and an intelligent agent based on convolution operation provided by an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a simulation tool and agent interaction framework of a simulation tool and agent interaction method based on convolution operation provided by an embodiment of the present invention;

[0056] Figure 3 A schematic flow chart of the convolution operation steps of a simulation tool for a method of interaction between a simulation tool and an intelligent agent based on a convolution operation provided by a certain embodiment of the present invention;

[0057] Figure 4 A schematic flow chart of the convolution operation steps of an intelligent agent in a method for interacting with an intelligent agent based on a convolution operation provided by a certain embodiment of the present invention;

[0058] Figure 5 A schematic diagram of the module structure of a simulation tool and an intelligent agent interaction system based on convolution operation provided by an embodiment of the present invention;

[0059] Figure 6 A schematic diagram of the structure of a convolution operation module of a simulation tool and an intelligent agent interaction system based on convolution operation is provided for a certain embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0061] Example 1

[0062] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of a method for interaction between a simulation tool and an intelligent agent based on convolution operation provided by an embodiment of the present invention. Figure 1 As shown, the embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on a convolution operation, including steps 101 to 105, each of which is specifically as follows:

[0063] Step 101, based on a plurality of time series data generated by the simulation tool, performing data analysis and data unification on the plurality of time series data to obtain a data set to be input;

[0064] As an example of this embodiment, based on several different simulation conditions, simulation is performed through the simulation tool to generate several time series data; dimension analysis is performed on the several time series data through the data analysis interface to obtain the several time series data of different dimensions; the several time series data of different dimensions are normalized to obtain the data set to be input. A specific implementation method is to obtain time series data simulated by a simulation tool, wherein the time series data includes: photovoltaic irradiance, load data and power generation data, etc.; the time series generated by the simulation tool is processed according to the parameter sample dimension; since the operation data of the new power system includes bus voltage value, phase angle and active power flow and reactive power flow of the transmission line, and each data has a different dimension, the data format obtained by the simulation tool is diverse, which depends on the design and application field of the simulation tool, such as: .CSV, .XML, .JSON, .NetCDF and .MAT files, etc., output files of BPA, PSASP and PSS / E, etc., which have fixed formats and are different from each other, and need to be parsed into data in a unified format. The embodiment of the present invention uses an API interface to parse the data and normalizes the parsed input data. The normalization generally adopts a minimum-maximum normalization method to eliminate the influence of different dimensions and ensure the consistency of subsequent analysis. The specific calculation method of the minimum-maximum normalization method shown is:

[0065]

[0066] Where: X i is the power system simulation state vector before data i is normalized; X i ' is the output value of the data normalization feature i after normalization, and its value is in the interval [0,1]; X i,max is the maximum value of the data before normalization; X i,min is the minimum value of the data before normalization, the state vector X i Including system frequency, node voltage, generator speed, load level, power flow and energy storage status, namely:

[0067] X i ={x1,x2,...,x n}

[0068] Through the above processing, the data set to be input is obtained, and the data set to be input is composed of the state vector X i ={x1,x2,...,x n}constitute.

[0069] Step 102, extracting features of the input data set by using a preset convolutional neural network model to obtain a first demand feature value;

[0070] As an example of this embodiment, the parameters of the corresponding preset convolutional neural network model are set according to the simulation requirements; wherein the preset convolutional neural network model includes: a convolution layer, a pooling layer and an output layer; based on the convolution layer, the feature of the input data set is extracted to obtain a feature map; based on the pooling layer, the feature map is reduced in dimension to obtain a pooled output value; based on the output layer, the probability distribution of the pooled output value is calculated by a preset activation function to obtain a first required feature value. A specific explanation is that by designing a convolutional neural network model, including: convolution layer, pooling layer and output layer, according to the characteristics of the power grid and simulation requirements, appropriate network parameters and layer structure are selected to effectively process and analyze complex data in the new power system. The convolutional neural network model supports multi-source heterogeneous data processing, and the convolution layer extracts key features in the operation of the power grid; the pooling layer is responsible for reducing the dimension of the feature map, reducing the computational complexity, and preventing overfitting; the output of the output layer is used as the input of the agent training, and the probability distribution is calculated according to the activation function (such as softmax); ensure that the feature data used has undergone a normalization preprocessing step, and divide the data into a training set and a validation set to monitor the model performance and prevent overfitting during the training process. The convolutional neural network model is used to extract features from complex new power system data to achieve network The forward propagation mechanism of the network is used to import the training data set into the network, adjust the structure and parameters of the convolutional neural network, ensure that the convolutional neural network can effectively extract useful features from the input data, and use the training data set to train the convolutional neural network model. After training, the model can accurately identify key dynamics in the power system, such as load fluctuations and instability of renewable energy generation, optimize the weights and parameters of the network, and use the validation set to monitor model performance; set parameters such as learning rate, batch size, and number of training rounds (Epochs), and select a suitable optimizer to help quickly and effectively adjust the weights and biases of the network, ensure the loss function and accuracy during training, ensure the forward optimization of the model, and make necessary model adjustments based on the performance during training and validation, change the network structure, adjust the parameters of the layer, or replace the activation function.

[0071] As another example of this embodiment, the data set to be input is input into the convolution layer of the preset convolutional neural network model; the convolution layer performs weight matrix operations on the data set to be input to obtain convolution eigenvalues; the convolution layer performs splicing, flattening and data image filling on the convolution eigenvalues ​​to obtain a feature map. In a specific implementation method, the convolution layer operates on time series data through a sliding convolution kernel for feature extraction of state vectors in a new power system to extract key features in power grid operation, such as load changes and power generation fluctuations. The feature map output by the convolution can effectively capture the instantaneous state and long-term trend of the new power system. After determining the number and configuration of the convolution layers, the extraction of local features of the input data is completed, and the eigenvalues ​​in the feature map are spliced ​​and flattened. The convolution operation is to perform weight matrix operations on the input data. The weight matrix is ​​called a convolution kernel. The formula for the convolution calculation is defined as follows:

[0072] Z i =σ(W ab LayerNorm(X i ,w k ))(i=1,2,……,n)

[0073] Where, X i is the state vector obtained by power system simulation, X i ={x1,x2,...,x n},w k is the power simulation system noise and external disturbance offset value; LayerNorm(·) is the layer normalization function, which performs normalization on the obtained state vector, W ab is the convolution kernel, which represents the state vector element of the a-th row and b-th column in the convolution kernel of the current layer; σ(·) is the activation function, which is used to introduce the nonlinear characteristics of the power system.

[0074] In order to ensure that the size of the feature map after convolution meets the expectation, it is necessary to add padding around the input image. The convolution operation will start from the upper left corner of the padded input data image and slide with a step size s each time. Suppose the input power system simulation data X i The size is n w ×n h ×a, the convolution kernel size is k×k, and the calculation formula for the feature map size after convolution is:

[0075]

[0076] Where T is the output size, n is the size of the input power system simulation data, and g is the padding size.

[0077] As another example of this embodiment, the feature map is input into the pooling layer of the preset convolutional neural network model; the feature map is subjected to maximum pooling and average pooling through the pooling layer to obtain a pooled output value. In one possible implementation method, the pooling layer is responsible for reducing the dimension of the feature map, reducing the computational complexity of the new power system data, and streamlining the features of the power system simulation data, enhancing the main features, and downplaying the secondary features. Usually, the pooling layer is divided into a maximum pooling layer and an average pooling layer, which are mainly used to perform secondary dimensionality reduction operations on the feature map, reduce the risk of overfitting during model training, and ensure the stability of the model under complex power system data. The lower pooling layer is generally located after the convolution layer, and its operation can be expressed as:

[0078] P i =f1[down(Z i *A)+l i ](i=1,2,……,m)

[0079] Where P i is the output of the i-th pooling layer; down(·) is the function of the next pooling layer; l i is the bias vector of the pooling layer under the i-th layer, A is the action vector, A={a1,a2,a3,...,a n},a n Represents the nth relevant decision quantity of the power system.

[0080] As another example of this embodiment, the pooled output is input to the output layer value of the preset convolutional neural network model; the preset activation function of the preset activation function enhancement model is used to perform probability distribution calculation on the pooled output value to obtain the first demand feature value. In one possible implementation method, the output layer uses the output of the pooling layer as the input of the output layer, applies the Leaky ReLU (·) activation function after each layer, and enhances the nonlinear expression ability of the model through the activation function; the output layer calculates the probability distribution according to the activation function (such as Leaky ReLU (·)) to ensure that the intelligent agent can make a quick and effective response based on the extracted features when making control decisions. Assuming that the activation function of the output layer is f2, the calculation formula of the fully connected layer is as follows:

[0081] R k =f2(Leaky ReLU(P i ),h k ),

[0082] Where P i is the output of the i-th pooling layer and the input of the output layer; Leaky ReLU(·) is the activation function, where h k is the weight vector.

[0083] Step 103, sending the first demand feature value to the agent to generate several training strategies;

[0084] As an example of this embodiment, the first requirement characteristic value is sent to the agent; based on different conditions and the task requirements preset by the agent, the first requirement characteristic value is trained on a strategy to generate several training strategies that match the task requirements preset by the agent under different conditions. Figure 3 , Figure 3 A schematic flow chart of a simulation tool convolution operation step of a simulation tool and an agent interaction method based on convolution operation provided by a certain embodiment of the present invention; Figure 3 As shown, through step 102, the convolutional neural network model is trained using the pre-set processed feature data (data set to be input), and the various data types and formats obtained by various simulation tools are pre-processed and transmitted to the convolutional neural network model for feature extraction. After the step flow, the data features required by the intelligent agent are obtained; then the corresponding feature values ​​are transmitted to the intelligent agent, and it is trained to generate several training strategies; the several strategies can be: load forecasting strategy, the intelligent agent extracts features from historical load data, meteorological information (such as temperature and humidity) and holiday effects. When the predicted load increases, the intelligent agent will correspondingly instruct the generator set to increase the corresponding power generation; power generation scheduling strategy, based on the extracted features, the intelligent agent can formulate a scheduling strategy to give priority to the use of low-cost renewable energy for power generation; for example, when the photovoltaic power generation is high, the intelligent agent will recommend reducing the use of coal-fired power generation, thereby reducing operating costs and carbon emissions; voltage control strategy, when a node voltage is lower than the set value, the intelligent agent generates a strategy to instruct the reactive compensation device to increase reactive power output and quickly increase the node voltage to a safe range; demand response strategy, during the peak period of power demand, the intelligent agent can formulate a dynamic pricing strategy to reasonably guide the user's power consumption behavior, thereby balancing the load and reducing the pressure of peak load on the power grid.

[0085] Step 104, based on the plurality of training strategies, matching corresponding simulation tool requirements, and generating a second requirement characteristic value;

[0086] As an example of this embodiment, based on a preset data annotation tool, the several training strategies are analyzed and annotated to generate probability distributions, actions and expected rewards corresponding to the several strategies; based on the simulation tool requirements, the probability distributions, actions and expected rewards corresponding to the several strategies are converted into a data format that can be analyzed by the simulation tool to obtain a second requirement characteristic value. One possible implementation method is described in detail in FIG. Figure 4 , Figure 4 A schematic flow chart of the convolution operation steps of an agent in a method for interacting with an agent based on a convolution operation is provided in accordance with an embodiment of the present invention; Figure 4As shown in the figure, after using the many strategies obtained by the agent training, the strategies are annotated and parsed, and the data format that can be used by the simulation tool is generated and passed to the simulation tool for simulation; the agent generates strategies by extracting features from the dynamic data of the new power system to adapt to different operating conditions and requirements. An interactive data annotation tool (Prodigy) can be used, which has a fast customization and iterative annotation mode to analyze and annotate the strategies generated by the agent; Among them, Prodigy allows users to quickly define annotation tasks and data flows according to specific task requirements, and adjust annotation rules and standards in combination with real-time feedback and strategy effects to accelerate the optimization process of the strategy; In addition, Prodigy Prodigy provides a visual interface that allows users to intuitively view and annotate data, and intuitively compare descriptions such as the output of generator sets before and after the action of the intelligent agent. This visualization capability can help power system operators and engineers quickly understand the decision-making process of the intelligent agent and make adjustments based on specific circumstances. For example, in voltage control strategies, users can intuitively see the effects of different strategies and make annotations and feedback based on this. It is worth mentioning that the use of a data annotation tool (Prodigy) can improve data annotation efficiency, allowing operators to quickly process large amounts of data, avoiding the inefficiency and tediousness of manual annotation in traditional methods, thereby saving time and labor costs.

[0087] Step 105: sending the second demand characteristic value to the simulation tool to complete an interaction between the simulation tool and the agent.

[0088] As an example of this embodiment, see Figure 2 , Figure 2 A schematic diagram of a simulation tool and agent interaction framework of a simulation tool and agent interaction method based on convolution operation provided by a certain embodiment of the present invention; Figure 2As shown, a simulation tool and intelligent agent interaction framework includes: multiple simulation tools, an interface based on convolution operation and an artificial intelligence training body; the interface based on convolution operation includes: a feature rule storage module, a strategy parsing module and a CNN model; the multiple simulation tools transmit data to the interface based on convolution operation through simulation and data preprocessing, extract key features of the data by extracting the feature rule storage module and using the CNN model and converting them into a data format acceptable to the intelligent agent, and sending the feature data to the artificial intelligence training body; after obtaining the feature data, the artificial intelligence training body generates several training strategies and sends them to the interface based on convolution operation through the data interface, and the interface based on convolution operation accesses the intelligent agent strategy through the strategy parsing module, returns the probability distribution, action and expected reward of the current strategy, etc., so as to parse the strategy, convert the strategy data into a data format acceptable to the simulation tool, and pass it to the simulation tool for simulation; completing an interaction between the simulation tool and the intelligent agent. In summary, multiple agents share information and coordinate strategies with multiple types of power grid simulation tools, adjust their parameters and structures according to the needs of the simulation environment, and adapt to different types of new power system configurations; the agents and simulation tools interact efficiently, extracting important information from the various data formats provided by various power grid simulation tools, and providing it to the agents for training to obtain strategies.

[0089] The embodiment of the present invention proposes a method for interaction between a simulation tool and an intelligent agent based on convolution operation. By acquiring the time series data generated by the simulation tool and parsing and unifying the event continuation data and the retrograde data, the nonlinear characteristics of the data and the dynamic changes of the power system under different environments are effectively acquired. At the same time, the data format is unified to reduce the operational limitations between the data during subsequent data processing, which helps to improve the interaction efficiency between the intelligent agent and the simulation tool. Then, through the convolution neural network model, the processed time series data is feature extracted to obtain the key features of the data to reduce the complexity of data processing, improve the data calculation efficiency and thus improve the interaction efficiency between the intelligent agent and the simulation tool. Then, the feature values ​​obtained by feature extraction are sent to the intelligent agent, so that the intelligent agent trains the corresponding training strategy, and the training strategy is converted into a data form acceptable to the simulation tool by matching the requirements of the simulation tool, and then the data is sent to the simulation tool for simulation. The above method is used to ensure the completion of an efficient data interaction. According to the characteristics of the power grid and the simulation requirements, the time series data is extracted and the key features of the data are obtained. The data is processed and the time series data is processed. The time series ... The key features of the data are extracted from the columns to effectively capture the instantaneous state and long-term trend of the power system, ensure that only valid data are calculated during the interaction process, reduce the transmission of redundant data, improve data computing power, and thus improve the interaction efficiency between the intelligent agent and the simulation tool; the dimension of the feature map is reduced through the pooling layer, the complexity of the operation and the streamlined simulation data features, the main features are strengthened, and the secondary features are weakened, and then the overfitting risk of model training is reduced through the maximum pooling and average pooling operations, and the operation stability of the model under complex data is improved; the nonlinear expression ability of the model is enhanced through the output layer through the activation function, ensuring that the intelligent agent can make a quick and effective response based on the extracted features when making control decisions, and improving the interaction efficiency between the intelligent agent and the simulation tool; multiple intelligent agents can interact with multiple simulation environments at the same time, and can adjust their parameters and structures according to the needs of the simulation environment to adapt to different types of new power system configurations; important information is extracted from the various data formats provided by various power grid simulation tools, and provided to the intelligent agent for training to obtain strategies, thereby improving the interaction efficiency between the intelligent agent and the simulation tool.

[0090] Example 2

[0091] See also Figure 5 , Figure 5 A schematic diagram of the module structure of a simulation tool and an intelligent agent interaction system based on convolution operation provided by an embodiment of the present invention. Figure 5 As shown, the embodiment of the present invention proposes a simulation tool and agent interaction system based on convolution operation, including:

[0092] Simulation tool input data processing module 501, convolution operation module 502, agent strategy generation module 503, strategy analysis module 504 and simulation tool data recovery module 505;

[0093] The simulation tool input data processing module 501 is used to perform data analysis and data unification on a number of time series data generated by the simulation tool to obtain a data set to be input;

[0094] The convolution operation module 502 is used to extract features of the to-be-input data set through a preset convolutional neural network model to obtain a first demand feature value;

[0095] For a specific example, see Figure 6 , Figure 6 A schematic diagram of the convolution operation module structure of a simulation tool and an intelligent agent interaction system based on convolution operation is provided in one embodiment of the present invention. Figure 6 As shown, the convolution operation module 502 is used to extract features of the input data set through a preset convolutional neural network model to obtain a first demand feature value, including:

[0096] Convolution model setting unit 601, convolution unit 602, pooling unit 603 and output unit 604;

[0097] The convolution model setting unit 601 is used to set the parameters of the corresponding preset convolution neural network model according to the simulation requirements; wherein the preset convolution neural network model includes: a convolution layer, a pooling layer and an output layer;

[0098] The convolution unit 602 is used to extract features of the input data set based on the convolution layer to obtain a feature map;

[0099] The pooling unit 603 is used to reduce the dimension of the feature map based on the pooling layer to obtain a pooling output value;

[0100] The output unit 604 is used to perform probability distribution calculation on the pooled output value based on the output layer by using a preset activation function to obtain a first demand feature value.

[0101] The agent strategy generation module 503 is used to send the first demand feature value to the agent to generate several training strategies;

[0102] The strategy analysis module 504 is used to match the corresponding simulation tool requirements based on the plurality of training strategies and generate a second requirement characteristic value;

[0103] The simulation tool data recovery module 505 is used to send the second demand characteristic value to the simulation tool to complete an interaction between the simulation tool and the agent.

[0104] The embodiment of the present invention proposes a simulation tool and intelligent agent interaction system based on convolution operation, which obtains the time series data generated by the simulation tool through the simulation tool input data processing module, analyzes the event power data and the retrograde data, and unifies the data, effectively obtains the nonlinear characteristics of the data and the dynamic changes of the power system under different environments, and unifies the data format to reduce the operational limitations between the data during subsequent data processing, which helps to improve the interaction efficiency between the intelligent agent and the simulation tool; then, through the convolution operation module, a convolutional neural network model is used to extract features from the processed time series data, obtain the key features of the data to reduce the complexity of data processing, improve the data calculation efficiency and thus improve the interaction efficiency between the intelligent agent and the simulation tool; then, through the intelligent agent strategy generation module, the feature value obtained by the feature extraction is sent to the intelligent agent, so that the intelligent agent trains the corresponding training strategy, and the strategy analysis module matches the needs of the simulation tool to convert the training strategy into a data form acceptable to the simulation tool, and then the data recovery module of the simulation tool sends the data to the simulation tool for simulation, and the above method is used to ensure the completion of an efficient data interaction.

[0105] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0106] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0107] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

Claims

1. A method for interaction between a simulation tool and an intelligent agent based on convolution operation, characterized in that: include: Based on the time series data generated by the simulation tool, performing data analysis and data unification on the time series data to obtain a data set to be input; Extracting features of the input data set by using a preset convolutional neural network model to obtain a first demand feature value; Sending the first demand feature value to the agent to generate a number of training strategies; Based on the plurality of training strategies, matching corresponding simulation tool requirements and generating a second requirement characteristic value; The second requirement characteristic value is sent to the simulation tool to complete an interaction between the simulation tool and the agent.

2. A method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 1, characterized in that: Based on the time series data generated by the simulation tool, the time series data are analyzed and unified to obtain a data set to be input, including: Based on a number of different simulation conditions, simulation is performed by the simulation tool to generate a number of time series data; Performing dimension analysis on the time series data through a data analysis interface to obtain the time series data of different dimensions; The time series data of different dimensions are normalized to obtain a data set to be input.

3. The method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 1, characterized in that: The method further comprises: extracting features of the input data set by using a preset convolutional neural network model to obtain a first demand feature value, including: The parameters of the corresponding preset convolutional neural network model are set according to the simulation requirements; wherein the preset convolutional neural network model includes: a convolution layer, a pooling layer and an output layer; Based on the convolutional layer, feature extraction is performed on the data set to be input to obtain a feature map; Based on the pooling layer, reducing the dimension of the feature map to obtain a pooling output value; Based on the output layer, a probability distribution calculation is performed on the pooled output value through a preset activation function to obtain a first demand feature value.

4. The method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 3, characterized in that: Based on the convolutional layer, feature extraction is performed on the data set to be input to obtain a feature map, including: Inputting the data set to be input into the convolution layer of the preset convolutional neural network model; Performing a weight matrix operation on the data set to be input through the convolution layer to obtain a convolution eigenvalue; The convolutional feature values ​​are concatenated, flattened and filled with data images through the convolution layer to obtain a feature map.

5. The method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 4, characterized in that: Based on the pooling layer, the feature map is reduced in dimension to obtain a pooling output value, including: Inputting the feature map into the pooling layer of the preset convolutional neural network model; The feature map is subjected to maximum pooling and average pooling through the pooling layer to obtain a pooling output value.

6. A method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 5, characterized in that: Based on the output layer, a probability distribution calculation is performed on the pooled output value by using a preset activation function to obtain a first demand feature value, including: Inputting the pooled output to the output layer value of the preset convolutional neural network model; The probability distribution of the pooled output value is calculated by using a preset activation function of the preset activation function enhancement model to obtain a first demand feature value.

7. The method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 1, characterized in that: The first demand feature value is sent to the agent to generate several training strategies, including: Sending the first demand feature value to the agent; Based on different conditions and preset task requirements of the intelligent agent, strategy training is performed on the first requirement characteristic value to generate several training strategies that match the preset task requirements of the intelligent agent under different conditions.

8. The method for interaction between a simulation tool and an intelligent agent based on convolution operation as claimed in claim 1, characterized in that: Based on the plurality of training strategies, the corresponding simulation tool requirements are matched to generate a second requirement characteristic value, including: Based on a preset data annotation tool, the training strategies are analyzed and annotated to generate probability distributions, actions, and expected rewards corresponding to the strategies; Based on the simulation tool requirements, the probability distributions, actions and expected rewards corresponding to the plurality of strategies are converted into a data format analyzable by the simulation tool to obtain a second requirement characteristic value.

9. A simulation tool and agent interaction system based on convolution operation, characterized in that: A method for interaction between a simulation tool and an intelligent agent based on convolution operation as described in any one of claims 1 to 8 comprises: Simulation tool input data processing module, convolution operation module, agent strategy generation module, strategy analysis module and simulation tool data recovery module; The simulation tool input data processing module is used to perform data analysis and data unification on the multiple time series data generated by the simulation tool to obtain the data set to be input; The convolution operation module is used to extract features of the to-be-input data set through a preset convolutional neural network model to obtain a first required feature value; The agent strategy generation module is used to send the first demand feature value to the agent to generate a number of training strategies; The strategy analysis module is used to match the corresponding simulation tool requirements based on the plurality of training strategies and generate a second requirement characteristic value; The simulation tool data recovery module is used to send the second demand characteristic value to the simulation tool to complete an interaction between the simulation tool and the agent.

10. A simulation tool and agent interaction system based on convolution operation as claimed in claim 9, characterized in that: The convolution operation module is used to extract features of the to-be-input data set through a preset convolutional neural network model to obtain a first required feature value, including: Convolutional model setup unit, convolution unit, pooling unit and output unit; The convolution model setting unit is used to set the parameters of the corresponding preset convolution neural network model according to the simulation requirements; wherein the preset convolution neural network model includes: a convolution layer, a pooling layer and an output layer; The convolution unit is used to extract features of the to-be-input data set based on the convolution layer to obtain a feature map; The pooling unit is used to reduce the dimension of the feature map based on the pooling layer to obtain a pooling output value; The output unit is used to perform probability distribution calculation on the pooled output value based on the output layer through a preset activation function to obtain a first demand feature value.

Images