Distribution network load transfer decision method and system

Abstract

The present invention discloses a method for making decisions on load transfer in a distribution network, comprising obtaining data information of a target distribution network; constructing a large language model of the distribution network for auxiliary decision-making on load transfer in the distribution network; constructing operation scenario data information of the target distribution network; constructing and training a load transfer model for the distribution network; obtaining data information of the target distribution network in real time, and using the large language model of the distribution network and the load transfer model of the distribution network to complete a load transfer decision for the target distribution network. The present invention also discloses a system for implementing the method for making decisions on load transfer in a distribution network. The present invention adopts a solution combining a large language model and deep reinforcement learning to make decisions on load transfer in the distribution network. Therefore, the present invention can not only implement decisions on load transfer in the distribution network, but also has higher reliability and better accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of electrical automation, and in particular relates to a load transfer decision method and system for a distribution network. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring a stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] As environmental issues become increasingly severe, more and more renewable energy generation systems are being integrated into the power system and generating electricity. The open, dynamic, and interactive nature of power system operation is becoming increasingly prominent. As a key factor in ensuring the safe and stable operation of the power system, load transfer solutions for distribution networks will become increasingly important.

[0004] Currently, load transfer plans for power distribution networks generally rely on pre-defined transfer rules. These rules are developed based on power system simulation data. When an abnormality in the power system requires load transfer, load transfer is performed according to the pre-defined rules. However, these load transfer plans are unable to cope with the extremely complex operating conditions of actual power systems.

[0005] Currently, researchers have proposed solutions based on artificial intelligence methods such as deep learning, reinforcement learning, and large models to assist in load transfer in distribution networks. However, these solutions only utilize the operating data of the power system and do not utilize the large amount of text data of the power system. Therefore, these solutions cannot fit the model according to the requirements of the regulation, resulting in poor reliability and accuracy of these solutions. Summary of the Invention

[0006] One of the objectives of the present invention is to provide a load transfer decision method for a distribution network with high reliability and good accuracy.

[0007] A second object of the present invention is to provide a system for implementing the distribution network load transfer decision-making method.

[0008] The load transfer decision method for a distribution network provided by the present invention comprises the following steps:

[0009] S1. Obtain data information of the target distribution network;

[0010] S2. Based on the data information obtained in step S1, a large language model of the distribution network is constructed to assist decision-making for load transfer in the distribution network;

[0011] S3. Based on the data information obtained in step S1, construct the target distribution network operation scenario data information;

[0012] S4. Build and train a distribution network load transfer model;

[0013] S5. Acquire data information of the target distribution network in real time, and use the distribution network large language model obtained in step S2 and the distribution network load transfer model obtained in step S4 to complete the load transfer decision of the target distribution network.

[0014] The step S1 of acquiring the target distribution network data information specifically includes the following steps:

[0015] Acquire text information of the target distribution network; the text information includes fault plan data information, emergency plan data information, maintenance plan data information and dispatching procedure text information;

[0016] Obtaining operating information of the target distribution network; the operating information includes distribution network topology information, distribution network power generation data information and distribution network load data information.

[0017] Step S2, based on the data information obtained in step S1, constructs a large language model of the distribution network, specifically including the following steps:

[0018] Extracting text data information: Extracting text information of fault plans, emergency plans, maintenance plans, and dispatching procedures. The text information of emergency plans includes plant and station names, fault causes, and emergency operation switch names. The text information of dispatching procedures includes plan preparation requirements and decision-making restrictions.

[0019] The extracted text data information is input into the large language model, and the prompt template is constructed using the layout;

[0020] Adding explanation and qualifier sentences to the large language model to improve the prompt words;

[0021] Finally, a large language model of the distribution network is constructed.

[0022] Step S3, based on the data information acquired in step S1, constructs the operation scenario data information of the target distribution network, specifically including the following steps:

[0023] Based on the acquired distribution network topology information, distribution network generation data information, and distribution network load data information, the physical node information of different components in the target distribution network is extracted to obtain a bus-node table: the physical nodes of all components are summarized and added to the bus table, and the buses and nodes are marked;

[0024] Connect different components according to physical nodes;

[0025] For power generation data and load data, missing data are filled through interpolation and state estimation algorithms;

[0026] Set the reward value of the distribution network load transfer model:

[0027] The large language model of the distribution network is adopted, and the concept of token is used to model the dispatching procedure text, which is expressed as

[0028]

[0029] Where Pr(t1,t2,...,t n ) is the generated regulation text model; n is the number of tokens used in this generation; t1~t n is a sequence of tokens obtained from the vocabulary by applying the probabilistic chain rule;

[0030] Using the large language model of the distribution network, a new text about rewards is generated based on the prompt words composed of the dispatching regulations text, and the generated new text is parsed into the reward value, which is expressed as

[0031] R(text)=parse{t1,t2,...,t n |a,s,c,q}

[0032] Where R(text) is the reward value generated by the dispatching procedure text; parse{} is the parsing function; a is the action of the distribution network load transfer model; s is the environment of the target distribution network system, c is the prompt word context, and q is the load transfer;

[0033] Finally, the reward value is expressed as

[0034] R epc =R f +R(text)-ΣwP

[0035] Where R epc is the total reward for this round of training, R f is the reward for completing power restoration, P is the penalty generated by the action, and w is the weight of the penalty.

[0036] The construction and training of the distribution network load transfer model described in step S4 specifically includes the following steps:

[0037] Based on the deep reinforcement learning model, a distribution network load transfer model is constructed;

[0038] The distribution network load transfer model reads environmental parameters; the environmental parameters include network topology, unit output, load power data and network component data;

[0039] The distribution network load transfer model randomly generates exploration actions;

[0040] Create a fault in the target distribution network and obtain the corresponding action output by the distribution network load transfer model;

[0041] Changing the state of the target distribution network according to the corresponding action output by the distribution network load transfer model;

[0042] Repeat the above steps until the training termination conditions are met; the training termination conditions include the completion of power supply restoration, voltage exceeding the limit, number of actions exceeding the limit, several consecutive identical actions, fault switch action or duration exceeding the limit;

[0043] The voltage over-limit value and load loss amount are calculated using the following formula:

[0044] V over =∑(V h.pu -V max.pu )+∑(V l.pu -V min.pu )

[0045] P loss =∑(P load -P gen )

[0046] Where V over The voltage exceeds the limit; V h.pu V is the per-unit value of the node voltage that exceeds the voltage upper limit; l.pu is the per-unit value of the node voltage below the voltage lower limit; V max.pu is the voltage upper limit; V min.pu is the voltage lower limit; P loss is the load loss; P load P is the load power of the electrical island that cannot fully restore power; gen is the power generation capacity;

[0047] The reward value is calculated based on the voltage limit and the load loss; the parameters of the distribution network load transfer model are changed according to the reward value; the network structure of the target distribution network is restored, the fault is re-created, and the next round of training is carried out.

[0048] The real-time target distribution network data information in step S5 uses the distribution network large language model obtained in step S2 and the distribution network load transfer model obtained in step S4 to complete the load transfer decision of the target distribution network, which specifically includes the following steps:

[0049] Obtain data information of the target distribution network in real time;

[0050] The data information obtained in step S1 is input into the constructed distribution network language model for judgment:

[0051] If the distribution network large language model determines that the input data information meets the scenario described by the set fault / maintenance plan and there is a corresponding load transfer plan, the load transfer plan is directly output to complete the load transfer decision of the target distribution network;

[0052] Otherwise, proceed to the next step;

[0053] The data information obtained in step S1 is input into the constructed distribution network load transfer model, and the data output by the distribution network load transfer model is obtained as the load transfer decision of the target distribution network.

[0054] The present invention also provides a system for implementing the distribution network load transfer decision method, comprising a data acquisition module, a model construction module, a data construction module, a model training module and a transfer decision module; the data acquisition module, the model construction module, the data construction module, the model training module and the transfer decision module are connected in series in sequence; the data acquisition module is used to acquire data information of the target distribution network, and upload the data information to the model construction module; the model construction module is used to construct a distribution network large language model based on the received data information, which is used for auxiliary decision-making on distribution network load transfer, and upload the data information to the data construction module; the data construction module is used to construct operation scenario data information of the target distribution network based on the received data information, and upload the data information to the model training module; the model training module is used to construct and train a distribution network load transfer model based on the received data information, and upload the data information to the transfer decision module; the transfer decision module is used to acquire data information of the target distribution network in real time based on the received data information, and use the obtained distribution network large language model and distribution network load transfer model to complete the load transfer decision of the target distribution network.

[0055] The distribution network load transfer decision-making method and system provided by the present invention adopts a solution combining a large language model and deep reinforcement learning to make decisions on distribution network load transfer. Therefore, the present invention can not only realize the decision-making of distribution network load transfer, but also has higher reliability and better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of the process flow of the present invention.

[0057] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0058] like Figure 1 The figure shows a flow chart of the method of the present invention: the load transfer decision method of the distribution network disclosed in the present invention comprises the following steps:

[0059] S1. Obtain data information of the target distribution network; specifically, the steps include:

[0060] Acquire text information of the target distribution network; the text information includes fault plan data information, emergency plan data information, maintenance plan data information and dispatching procedure text information;

[0061] Obtaining operating information of the target distribution network; the operating information includes distribution network topology information, distribution network power generation data information, and distribution network load data information;

[0062] S2. Based on the data information obtained in step S1, a large language model of the distribution network is constructed to assist decision-making on load transfer in the distribution network; specifically, the steps include:

[0063] Extracting text data information: Extracting text information of fault plans, emergency plans, maintenance plans, and dispatching procedures. The text information of emergency plans includes plant and station names, fault causes, and emergency operation switch names. The text information of dispatching procedures includes plan preparation requirements and decision-making restrictions.

[0064] The extracted text data information is input into the large language model, and the prompt template is constructed using the layout;

[0065] Adding explanation and qualifier sentences to the large language model to improve the prompt words;

[0066] Finally, a large language model of the distribution network is constructed;

[0067] S3. Based on the data information obtained in step S1, construct the operation scenario data information of the target distribution network; specifically comprising the following steps:

[0068] Based on the obtained distribution network topology information, distribution network generation data information, and distribution network load data information (which can be obtained from the power system D5000 dispatching system), the physical node information of different components in the target distribution network is extracted to obtain a bus-node table: the physical nodes of all components are summarized and added to the bus table, and the buses and nodes are marked;

[0069] Connect different components according to physical nodes;

[0070] For power generation and load data, missing data is supplemented through interpolation and state estimation algorithms. If the power data of a generator or load is missing, it can be interpolated through the data of the previous and next time nodes, estimated through current measurement data, or estimated through state estimation methods.

[0071] Set the reward value of the distribution network load transfer model:

[0072] The large language model of the distribution network is adopted, and the concept of token is used to model the dispatching procedure text, which is expressed as

[0073]

[0074] Where Pr(t1,t2,...,t n ) is the generated regulation text model; n is the number of tokens used in this generation; t1~t n is a sequence of tokens obtained from the vocabulary by applying the probabilistic chain rule;

[0075] Using the large language model of the distribution network, a new text about rewards is generated based on the prompt words composed of the dispatching regulations text, and the generated new text is parsed into the reward value, which is expressed as

[0076] R(text)=parse{t1,t2,...,t n |a,s,c,q}

[0077] Where R(text) is the reward value generated by the dispatching procedure text; parse{} is the parsing function; a is the action of the distribution network load transfer model; s is the environment of the target distribution network system, c is the prompt word context, and q is the load transfer;

[0078] Finally, the reward value is expressed as

[0079] R epc =R f +R(text)-∑wP

[0080] Where R epc is the total reward for this round of training, R f is the reward for completing power restoration, P is the penalty generated by the action, and w is the weight of the penalty;

[0081] S4. Build and train a distribution network load transfer model; specifically, the steps include:

[0082] Based on the deep reinforcement learning model, a distribution network load transfer model is constructed;

[0083] The distribution network load transfer model reads environmental parameters; the environmental parameters include network topology, unit output, load power data and network component data;

[0084] The distribution network load transfer model randomly generates exploration actions;

[0085] Create a fault in the target distribution network and obtain the corresponding action output by the distribution network load transfer model;

[0086] Changing the state of the target distribution network according to the corresponding action output by the distribution network load transfer model;

[0087] Repeat the above steps until the training termination conditions are met; the training termination conditions include the completion of power supply restoration, voltage exceeding the limit, number of actions exceeding the limit, several consecutive identical actions, fault switch action or duration exceeding the limit;

[0088] The voltage over-limit value and load loss amount are calculated using the following formula:

[0089] V over =∑(V h.pu -V max.pu )+Σ(V l.pu -V min.pu )

[0090] P loss =∑(P load -P gen )

[0091] Where V over The voltage exceeds the limit; V h.pu V is the per-unit value of the node voltage that exceeds the voltage upper limit; l.pu is the per-unit value of the node voltage below the voltage lower limit; V max.pu is the voltage upper limit; V min.pu is the voltage lower limit; P loss is the load loss; P load P is the load power of the electrical island that cannot fully restore power; gen is the power generation capacity;

[0092] The reward value is calculated based on the voltage over-limit value and the load loss amount; the parameters of the distribution network load transfer model are changed according to the reward value; the network structure of the target distribution network is restored, the fault is re-created, and the next round of training is carried out;

[0093] S5 real-time acquisition of target distribution network data information, using the distribution network large language model obtained in step S2 and the distribution network load transfer model obtained in step S4, complete the target distribution network load transfer decision; specifically comprising the following steps:

[0094] Obtain data information of the target distribution network in real time;

[0095] The data information obtained in step S1 is input into the constructed distribution network language model for judgment:

[0096] If the distribution network large language model determines that the input data information meets the scenario described by the set fault / maintenance plan and there is a corresponding load transfer plan, the load transfer plan is directly output to complete the load transfer decision of the target distribution network;

[0097] Otherwise, proceed to the next step;

[0098] The data information obtained in step S1 is input into the constructed distribution network load transfer model, and the data output by the distribution network load transfer model is obtained as the load transfer decision of the target distribution network.

[0099] like Figure 2 The figure shows a schematic diagram of the functional modules of the system of the present invention: the system disclosed in the present invention for realizing the distribution network load transfer decision method includes a data acquisition module, a model construction module, a data construction module, a model training module and a transfer decision module; the data acquisition module, the model construction module, the data construction module, the model training module and the transfer decision module are connected in series in sequence; the data acquisition module is used to acquire data information of the target distribution network and upload the data information to the model construction module; the model construction module is used to construct a large language model of the distribution network based on the received data information, which is used for auxiliary decision-making of the distribution network load transfer, and upload the data information to the data construction module; the data construction module is used to construct operation scenario data information of the target distribution network based on the received data information, and upload the data information to the model training module; the model training module is used to construct and train a distribution network load transfer model based on the received data information, and upload the data information to the transfer decision module; the transfer decision module is used to acquire data information of the target distribution network in real time based on the received data information, and use the obtained distribution network large language model and distribution network load transfer model to complete the load transfer decision of the target distribution network.

Claims

1. A method for load transfer decision-making in a distribution network, comprising the following steps: S1. Obtain data information of the target distribution network; specifically, the steps include: Acquire text information of the target distribution network; the text information includes fault plan data information, emergency plan data information, maintenance plan data information and dispatching procedure text information; Obtaining operating information of the target distribution network; the operating information includes distribution network topology information, distribution network power generation data information, and distribution network load data information; S2. Based on the data information obtained in step S1, a large language model of the distribution network is constructed to assist decision-making for load transfer in the distribution network; S3. Based on the data information obtained in step S1, construct the target distribution network operation scenario data information; S4. Build and train a distribution network load transfer model; S5 real-time acquisition of target distribution network data information, using the distribution network large language model obtained in step S2 and the distribution network load transfer model obtained in step S4, complete the target distribution network load transfer decision; specifically comprising the following steps: Obtain data information of the target distribution network in real time; The data information obtained in step S1 is input into the constructed distribution network language model for judgment: If the distribution network large language model determines that the input data information meets the scenario described by the set fault / maintenance plan and there is a corresponding load transfer plan, the load transfer plan is directly output to complete the load transfer decision of the target distribution network; Otherwise, proceed to the next step; The data information obtained in step S1 is input into the constructed distribution network load transfer model, and the data output by the distribution network load transfer model is obtained as the load transfer decision of the target distribution network.

2. The load transfer decision method for distribution network according to claim 1 is characterized in that Step S2, based on the data information obtained in step S1, constructs a large language model of the distribution network, specifically including the following steps: Extracting text data information: Extracting text information of fault plans, emergency plans, maintenance plans, and dispatching procedures. The text information of emergency plans includes plant and station names, fault causes, and emergency operation switch names. The text information of dispatching procedures includes plan preparation requirements and decision-making restrictions. The extracted text data information is input into the large language model, and the prompt template is constructed using the layout; Adding explanation and qualifier sentences to the large language model to improve the prompt words; Finally, a large language model of the distribution network is constructed.

3. The load transfer decision method for distribution network according to claim 2, characterized in that Step S3, based on the data information acquired in step S1, constructs the operation scenario data information of the target distribution network, specifically including the following steps: Based on the acquired distribution network topology information, distribution network generation data information, and distribution network load data information, the physical node information of different components in the target distribution network is extracted to obtain a bus-node table: the physical nodes of all components are summarized and added to the bus table, and the buses and nodes are marked; Connect different components according to physical nodes; For power generation data and load data, missing data are filled through interpolation and state estimation algorithms; Set the reward value of the distribution network load transfer model: The large language model of the distribution network is adopted, and the concept of token is used to model the dispatching procedure text, which is expressed as Where Pr(t1,t2,...,t n ) is the generated regulation text model; n is the number of tokens used in this generation; t1~t n is a sequence of tokens obtained from the vocabulary by applying the probabilistic chain rule; Using the large language model of the distribution network, a new text about rewards is generated based on the prompt words composed of the dispatching regulations text, and the generated new text is parsed into the reward value, which is expressed as R(text)=parse{t1,t2,...,t n |a,s,c,q} Where R(text) is the reward value generated by the dispatching procedure text; parse{} is the parsing function; a is the action of the distribution network load transfer model; s is the environment of the target distribution network system, c is the prompt word context, and q is the load transfer; Finally, the reward value is expressed as R epc =R f +R(text)-∑wP Where R epc is the total reward for this round of training, R f is the reward for completing power restoration, P is the penalty generated by the action, and w is the weight of the penalty.

4. The load transfer decision method for distribution network according to claim 3 is characterized in that The construction and training of the distribution network load transfer model described in step S4 specifically includes the following steps: Based on the deep reinforcement learning model, a distribution network load transfer model is constructed; The distribution network load transfer model reads environmental parameters; the environmental parameters include network topology, unit output, load power data and network component data; The distribution network load transfer model randomly generates exploration actions; Create a fault in the target distribution network and obtain the corresponding action output by the distribution network load transfer model; Changing the state of the target distribution network according to the corresponding action output by the distribution network load transfer model; Repeat the above steps until the training termination conditions are met; Conditions for training termination include power supply restoration, voltage exceeding the limit, number of actions exceeding the limit, consecutive identical actions, fault switch action or duration exceeding the limit; The voltage over-limit value and load loss amount are calculated using the following formula: V over =∑(V h.pu -V max.pu )+∑(V l.pu -V min.pu ) P loss =∑(P load -P gen ) Where V over The voltage exceeds the limit; V h.pu V is the per-unit value of the node voltage that exceeds the voltage upper limit; l.pu is the per-unit value of the node voltage below the voltage lower limit; V max.pu is the voltage upper limit; V min.pu is the voltage lower limit; P loss is the load loss; P load P is the load power of the electrical island that cannot fully restore power; gen is the power generation capacity; The reward value is calculated based on the voltage over-limit value and the load loss amount; Changing the parameters of the load transfer model of the distribution network according to the reward value; Restore the network structure of the target distribution network, recreate the fault, and conduct the next round of training.

5. A system for implementing the load transfer decision method of a distribution network according to any one of claims 1 to 4, characterized in that It includes a data acquisition module, a model construction module, a data construction module, a model training module and a transfer decision module; the data acquisition module, the model construction module, the data construction module, the model training module and the transfer decision module are connected in series in sequence; the data acquisition module is used to obtain data information of the target distribution network and upload the data information to the model construction module; the model construction module is used to construct a large language model of the distribution network based on the received data information, which is used for auxiliary decision-making on the load transfer of the distribution network, and upload the data information to the data construction module; The data construction module is used to construct the operation scenario data information of the target distribution network based on the received data information, and upload the data information to the model training module; The model training module is used to construct and train the distribution network load transfer model based on the received data information, and upload the data information to the transfer decision module; The transfer decision module is used to obtain the data information of the target distribution network in real time based on the received data information, and use the obtained distribution network large language model and distribution network load transfer model to complete the load transfer decision of the target distribution network.

Images