Power supply station business indicator early warning method, system, storage medium, and electronic terminal

By converting the power system environment into a reinforced learning environment, and using automatic coding machines and dual-delay depth deterministic strategy gradient algorithm to build an agent, the problem of low efficiency of traditional early warning methods is solved, and fast and accurate abnormal situation capture and early warning are achieved. Improves the management efficiency and safety and stability of the power system.

CN118572895BActive Publication Date: 2024-11-05STATE GRID TIANJIN ELECTRIC POWER COMPANY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411018272.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-05
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The business indicator early warning method of traditional power supply stations is based on rules and experience, and it is difficult to adapt to complex and changeable power system situations, resulting in low efficiency when processing large amounts of data, and the inability to timely detect and warn of new faults or abnormal situations, affecting the safe and stable operation of the power system.

Method used

Transform the power system environment into a reinforced learning environment, collect and preprocess historical data, and build an agent using improved automatic encoding machines and dual-delay depth deterministic strategy gradient algorithms to achieve real-time monitoring and early warning.

Benefits of technology

It significantly improves intelligent early warning capabilities, quickly capture abnormal situations, shortens early warning response time, improves management efficiency, adapts to changes in the power system, and ensures the safe and stable operation of the power system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118572895B_ABST
    Figure CN118572895B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, system, storage medium and electronic terminal for early warning of business indicators of power supply stations. The method comprises the following steps: converting the real power system environment into a reinforcement learning environment capable of interacting with an intelligent agent; collecting historical data related to business indicators of power supply stations and preprocessing them; filtering the preprocessed historical data using an improved automatic encoder to obtain filtered historical data; constructing an intelligent agent based on a double-delayed deep deterministic policy gradient algorithm based on the filtered historical data; collecting data related to business indicators of power supply stations in real time, using the intelligent agent based on the double-delayed deep deterministic policy gradient algorithm to monitor the data related to business indicators of power supply stations collected in real time, and issuing an early warning according to the early warning strategy. The use of this invention effectively reduces the impact of abnormal business indicators of power supply stations on the safe and stable operation of power systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power business of power supply stations, and in particular to a power supply station business indicator early warning method, system, storage medium and electronic terminal. Background Art

[0002] The safe and stable operation of the power system is crucial to the power supply and production of modern society. However, faults and abnormal conditions may occur in the power system, such as power equipment failures, sudden changes in power load, etc. Traditional power supply station business indicator early warning methods are often based on rules or experience, relying on manually formulated rules and thresholds to determine whether to issue an early warning. These methods have limitations, are difficult to adapt to complex and changeable power system conditions, and are inefficient when processing large amounts of data. Once a new type of fault or abnormal condition occurs in the power system, traditional methods often fail to detect and warn in a timely manner, resulting in untimely response from the power supply station, affecting the safe and stable operation of the power supply system in the substation area. Summary of the invention

[0003] In view of the technical problems pointed out in the background technology, the purpose of the present invention is to provide a power supply station business indicator early warning method, system and storage medium, and electronic terminal.

[0004] To achieve the purpose of the present invention, the technical solution provided by the present invention is specifically as follows:

[0005] First aspect

[0006] The present invention provides a power supply station business indicator early warning method, comprising the following steps:

[0007] Step S1: transform the real power system environment into a reinforcement learning environment that can interact with the intelligent agent;

[0008] Step S2: Collect historical data related to the business indicators of the power supply station and perform preprocessing;

[0009] Step S3: screening the preprocessed historical data using an improved automatic encoding machine to obtain screened historical data;

[0010] Step S4: Based on the filtered historical data, construct an agent based on the double-delayed deep deterministic policy gradient algorithm;

[0011] Step S5: Collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

[0012] Second aspect

[0013] Corresponding to the above method, the present invention also provides a power supply station business indicator early warning system, comprising the following units: a reinforcement learning environment construction unit, a historical data collection unit, a historical data screening unit, an intelligent body construction unit, and an early warning unit;

[0014] The reinforcement learning environment construction unit is used to transform the real power system environment into a reinforcement learning environment capable of interacting with the intelligent agent;

[0015] The historical data collection unit is used to collect historical data related to the business indicators of the power supply station and perform preprocessing;

[0016] The historical data screening unit is used to screen the pre-processed historical data using an improved automatic encoding machine to obtain screened historical data;

[0017] The intelligent agent construction unit is used to construct an intelligent agent based on a double-delayed deep deterministic policy gradient algorithm based on the filtered historical data;

[0018] The early warning unit is used to collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

[0019] The third aspect

[0020] The present invention also provides a storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the power supply station business indicator early warning method.

[0021] The fourth aspect

[0022] The present invention also provides a terminal, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the power supply station business indicator early warning method.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] (1) The intelligent early warning capability is significantly improved. The invention uses deep learning and automatic coding technology to enable the intelligent agent to accurately perceive and analyze the complex characteristics of the power supply station's business indicators. Through the construction of a double-delay neural network, the intelligent agent can quickly and accurately capture abnormal situations from the power system environment, achieving a significant improvement in the intelligent early warning capability, which will greatly shorten the warning response time and effectively reduce the impact of abnormal business indicators of the power supply station on the safe and stable operation of the power system.

[0025] (2) Obvious advantages in adaptability and flexibility. The adoption of the dual-delay deep deterministic policy gradient algorithm enables the intelligent agent to have good adaptability and flexibility. Through continuous optimization training, the intelligent agent can continuously adapt to the changes and evolution of the power system, thereby coping with the ever-changing business indicators of the power supply station. This enables the system to quickly adapt to new faults or abnormal situations and maintain efficient and accurate early warning and emergency response capabilities.

[0026] (3) Improve the management efficiency of power supply stations. The intelligent early warning method of this invention can realize comprehensive monitoring and early warning of various business indicators, effectively improving the management efficiency of power supply stations. Through the application of automatic coding technology, the intelligent agent compresses and extracts features from the input data, reduces the data dimension, reduces the workload of data processing, and helps to improve the efficiency of business indicator early warning and emergency response of power supply stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of the method flow provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0029] The power supply station business indicator early warning method, system, storage medium, and electronic terminal proposed in the present invention are a power supply station business indicator early warning solution based on the automatic encoding double-delay deep deterministic policy gradient algorithm. This solution combines the advantages of automatic encoding and double-delay deep deterministic policy gradient algorithms to achieve the extraction of abnormal patterns of business indicators and the optimization of early warning decisions. The business indicators are modeled and predicted by deep learning technology, and the early warning decisions are optimized by combining reinforcement learning technology, which improves the accuracy and efficiency of the power supply station business indicator early warning and ensures the safe and stable operation of the power supply system in the power supply station area. The background technology of this invention provides strong support for the development of power supply station business indicator early warning methods, and has broad application prospects and commercial value.

[0030] Example 1

[0031] like Figure 1As shown, this embodiment provides a power supply station business indicator early warning method, including the following steps:

[0032] Step S1: Convert the real power system environment into a reinforcement learning environment that can interact with the intelligent agent, including defining environmental parameters such as state space and action space.

[0033] The detailed steps to build a reinforcement learning environment are as follows:

[0034] Step S1.1: Define the state space. The state space is the state representation of the power system environment, which includes the key information of the power system. In the power system, the state quantities that need to be observed may include but are not limited to:

[0035] Power load: includes the current load conditions of each node in the power system, that is, the power consumption conditions.

[0036] Energy supply situation: including the current supply situation of various energy sources in the power system, such as traditional thermal power, renewable energy (wind power, photovoltaics, etc.), etc.

[0037] Equipment operating status: including the operating status of each device in the power system (such as transformers, generators, etc.), whether it is operating normally, etc.

[0038] Substation power supply topology: includes the connection status between each node in the power system, which constitutes the power supply system topology.

[0039] Step S1.2: Define the action space. In the scenario of power supply station business indicator warning, the action space may include but is not limited to:

[0040] Alarm threshold adjustment: Dynamically adjust the alarm threshold of the power supply station's business indicators to make the early warning system more sensitive or more relaxed to meet the early warning needs in different situations.

[0041] Data collection frequency adjustment: Dynamically adjust the collection frequency of power supply station business indicator data to balance the cost of data collection and the ability to perceive abnormal situations in a timely manner.

[0042] Selection of data preprocessing methods: According to the learning effect of the intelligent agent and the characteristics of the power supply station's business indicators, select appropriate data preprocessing methods, such as missing value filling algorithm, outlier processing algorithm, etc.

[0043] Alarm strategy priority setting: Set the priority of different warning strategies to ensure that when multiple warning situations occur at the same time, they can be responded and processed according to priority.

[0044] Warning notification method: Select an appropriate warning notification method, such as SMS, email, phone, etc., to notify relevant personnel of the warning information in a timely manner.

[0045] Optimization of early warning processing flow: According to actual operation conditions and business needs, optimize the early warning processing flow to improve the efficiency and accuracy of early warning response.

[0046] The definition of the above action space can be adjusted and expanded according to the needs and actual conditions of the specific power supply station business indicator early warning system to achieve a more intelligent and flexible early warning and response mechanism.

[0047] Step S2: Collect historical data related to the business indicators of the power supply station and perform preprocessing;

[0048] Among them, the historical data related to the business indicators of the power supply station include the number of smart meters, the number of non-smart meters, the number of negative loss areas, the number of high loss areas, the power supply, the completion rate of line loss work orders, the success rate of line loss management, the number of repeatedly cut and changed areas, the completion rate of collection work orders, the recovery rate of overdue electricity bills, the meter reading problem rate, the cumulative line loss rate, the cumulative line loss qualified rate, the average monthly collection success rate, the proportion of all collection areas, the online monitoring rate, the recovery rate of electricity bills in the previous month, the recovery rate of electricity bills in the current year, the amount of arrears in the current year, the amount of arrears in the monthly arrears, The amount of old electricity bills owed, the cumulative amount of arrears, the amount of liquidated damages for the year, the number of overdue work orders for residential business, the number of overdue work orders for non-residential business, the daily line loss qualification rate, the daily line loss rate, the daily data collection success rate, the number of users whose data collection is unsuccessful, the daily distribution transformer load qualification rate, the daily heavy load number, the daily overload number, the daily low voltage number, the electricity bill recovery rate for the current month, the number of users in arrears for the current month, the amount of arrears for the current month, the number of substations, the number of users, the number of low-voltage non-residential households, the number of low-voltage residential households, the number of high-voltage users, the number of substations, the number of lines and the number of transformers, etc.

[0049] It should be noted that the collected historical data related to the business indicators of the power supply station are preprocessed, including missing value filling, outlier processing, etc.

[0050] Step S3: screening the preprocessed historical data using an improved automatic encoding machine to obtain screened historical data;

[0051] It should be noted that the improved autoencoder is used to encode and screen the preprocessed historical data. The autoencoder is an unsupervised learning neural network model that achieves data dimensionality reduction and feature extraction through encoding and decoding processes. After extracting key features from the preprocessed historical data of the power system, these features are used as input to provide condensed information for the subsequent deep reinforcement learning algorithm, helping the intelligent agent to play a better role in power system fault detection and emergency response.

[0052] The training of the improved automatic encoder is described as follows:

[0053] h1=σ e (W1x+b1) (1)

[0054] z=σ d (W2h1+b2) (2)

[0055] Wherein, W1, b1 are the weight and bias of the encoding, W2, b2 are the weight and bias of the encoding, x is the input historical data, z is the encoding result, and the encoding result is the filtered historical data; h1 is the hidden layer variable, σ e , σ d are all constants;

[0056] The loss function of the improved automatic encoder is described as follows:

[0057]

[0058] Among them, J(·) is the total loss function and L(·) is the sub-loss function.

[0059] Step S4: Based on the filtered historical data, construct an agent based on the double-delayed deep deterministic policy gradient algorithm;

[0060] The agent is a deep reinforcement learning algorithm, which is particularly suitable for solving problems in continuous action space and high-dimensional state space. Its core is a double-delay neural network, which consists of two target networks Q1(S,A), Q2(S,A) and a policy network μ(s).

[0061] The execution process of the double-delay neural network is as follows:

[0062] For the current state S, S = z, perform action A:

[0063] A=μ(s)+N (4)

[0064] Where N is the random noise in the noise model; perform action A, get reward R and next state S', store the experience (S, A, R, S') in the experience buffer; randomly extract M experiences (S i ,A i ,R i ,S' i ), if S' i is the terminal state, R i Assign to the value function.

[0065] The training process of the double-delay neural network is as follows:

[0066] The value function is the experience reward R iThe sum of the discounted future rewards. To calculate the cumulative reward, the agent first calculates the next action by passing the next state S' in the sampled experience to the target Actor, and adds noise to the action. The agent passes the next action to the target Critic to obtain the cumulative reward. In each training step, the Critic parameters are updated by minimizing the loss L of all sampled experiences:

[0067]

[0068] Among them, L k is the calculated loss, y i is the expected decision, Q k is the reinforcement learning weight factor, are the learnable parameters involved, and i represents the i-th experience.

[0069] In addition, this application innovatively designs a reasonable reward function for the agent, which comprehensively considers indicators such as the safety, stability and efficiency of the power system. The reward function can be expressed as:

[0070] R(S,A,S')=α·Safe(S,A,S')+β·Stab(S,A,S')-τ·Delay(S,A,S') (6)

[0071] Among them, S represents the current state, A represents the agent action, S' represents the next state, Safe(S,A,S') is the safety reward, Stab(S,A,S') is the stability reward, Delay(S,A,S') is the delay penalty, α, β, τ are adjustment factors used to balance the importance of each indicator.

[0072] The present invention designs a reasonable reward function to guide the agent to gradually learn the optimal early warning strategy for power supply station business indicators. Through multiple optimization trainings, the agent continuously iterates and updates the strategy, enabling it to make accurate and reliable early warning and emergency response decisions when dealing with different situations. This makes the early warning results more accurate and reliable, and helps to improve the operating efficiency and stability of the power system.

[0073] The present invention innovatively uses a dual-delay neural network agent to perceive the state and make decisions and actions from a complex power system through its excellent environmental perception ability, and further guides the agent to gradually learn the best decision-making strategy through an innovative reward function. Finally, the real-time monitoring module is used to timely adjust the operating state of the system and the learning mode of the agent.

[0074] Step S5: Collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

[0075] It should be noted that the intelligent agent is embedded in the power supply station business indicator early warning and response system. The business indicator data of the power supply station is monitored in real time (the data will also be pre-processed as in step S2 and screened as in step S3), and intelligent early warning and emergency response are performed according to the early warning strategy of the intelligent agent. When the business indicators of the power supply station are abnormal, the intelligent agent quickly takes emergency response measures to ensure the safe and stable operation of the power system.

[0076] Example 2

[0077] Corresponding to the above method, this embodiment provides a power supply station business indicator early warning system, including the following units: a reinforcement learning environment construction unit, a historical data collection unit, a historical data screening unit, an intelligent agent construction unit, and an early warning unit;

[0078] The reinforcement learning environment construction unit is used to transform the real power system environment into a reinforcement learning environment capable of interacting with the intelligent agent;

[0079] The historical data collection unit is used to collect historical data related to the business indicators of the power supply station and perform preprocessing;

[0080] The historical data screening unit is used to screen the pre-processed historical data using an improved automatic encoding machine to obtain screened historical data;

[0081] The intelligent agent construction unit is used to construct an intelligent agent based on a double-delayed deep deterministic policy gradient algorithm based on the filtered historical data;

[0082] The early warning unit is used to collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

[0083] Among them, the historical data related to the business indicators of the power supply station include the number of smart meters, the number of non-smart meters, the number of negative loss areas, the number of high loss areas, the power supply, the completion rate of line loss work orders, the success rate of line loss management, the number of repeatedly cut and changed areas, the completion rate of collection work orders, the recovery rate of overdue electricity bills, the meter reading problem rate, the cumulative line loss rate, the cumulative line loss qualified rate, the average monthly collection success rate, the proportion of all collection areas, the online monitoring rate, the recovery rate of electricity bills in the previous month, the recovery rate of electricity bills in the current year, the amount of arrears in the current year, the amount of arrears in the monthly arrears, The amount of old electricity bills owed, the cumulative amount of arrears, the amount of liquidated damages for the year, the number of overdue work orders for residential business, the number of overdue work orders for non-residential business, the daily line loss qualification rate, the daily line loss rate, the daily data collection success rate, the number of users whose data collection is unsuccessful, the daily distribution transformer load qualification rate, the daily heavy load number, the daily overload number, the daily low voltage number, the electricity bill recovery rate for the current month, the number of users in arrears for the current month, the amount of arrears for the current month, the number of substations, the number of users, the number of low-voltage non-residential households, the number of low-voltage residential households, the number of high-voltage users, the number of substations, the number of lines and the number of transformers, etc.

[0084] The training of the improved automatic encoder is described as follows:

[0085] h1=σ e (W1x+b1) (1)

[0086] z=σ d (W2h1+b2) (2)

[0087] Wherein, W1, b1 are the weight and bias of the encoding, W2, b2 are the weight and bias of the encoding, x is the input historical data, z is the encoding result, and the encoding result is the filtered historical data; h1 is the hidden layer variable, σ e , σ d are all constants;

[0088] The loss function of the improved automatic encoder is described as follows:

[0089]

[0090] Among them, J(·) is the total loss function and L(·) is the sub-loss function.

[0091] Among them, the core of the intelligent agent based on the double-delay deep deterministic policy gradient algorithm is a double-delay neural network, including two target networks and one policy network.

[0092] Example 3

[0093] This embodiment provides a storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the power supply station business indicator early warning method.

[0094] Example 4

[0095] This embodiment provides an electronic terminal, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the power supply station business indicator early warning method.

[0096] The optional implementation modes of the embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation modes. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical scheme of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention.

Claims

1. A power supply station business indicator early warning method, characterized in that: The steps include: Step S1: transform the real power system environment into a reinforcement learning environment that can interact with the intelligent agent; Step S2: Collect historical data related to the business indicators of the power supply station and perform preprocessing; Step S3: screening the preprocessed historical data using an improved automatic encoding machine to obtain screened historical data; The training of the improved autoencoder is described as follows: h1=σ e (W1x+b1) (1) z=σ d (W2h1+b2) (2) Wherein, W1, b1 are the weight and bias of the encoding, W2, b2 are the weight and bias of the encoding, x is the input historical data, z is the encoding result, and the encoding result is the filtered historical data; h1 is the hidden layer variable, σ e , σ d are all constants; The loss function of the improved automatic encoder is described as follows: Among them, J(·) is the total loss function, L(·) is the sub-loss function; Step S4: Based on the filtered historical data, construct an agent based on the double-delayed deep deterministic policy gradient algorithm; The core of the agent is a dual-delay neural network, which consists of two target networks Q1(S, A), Q2(S, A) and a policy network μ(s); The execution process of the double-delay neural network is as follows: For the current state S, S = z, perform action A: A=μ(s)+N (4) Where N is the random noise in the noise model; perform action A, get reward R and next state S′, store the experience (S, A, R, S′) in the experience buffer; randomly extract M experiences (S i , A i , R i , S′ i ), if S′ i is the terminal state, R i Assign to the value function; The training process of the double-delay neural network is as follows: The value function is the experience reward R i The agent first calculates the next action by passing the next state S′ in the sampled experience to the target Actor and adds noise to the action. The agent passes the next action to the target Critic to obtain the cumulative reward. In each training step, the Critic parameters are updated by minimizing the loss L of all sampled experiences: Among them, L k is the calculated loss, y i is the expected decision, Q k is the reinforcement learning weight factor, is the learnable parameter involved, i represents the i-th experience; The agent’s reward function is expressed as: R(S,A,S′)=α·Safe(S,A,S′)+β·Stab(S,A,S′)-τ·Delay(S,A,S′) (6) Among them, S represents the current state, A represents the agent action, S′ represents the next state, Safe(S, A, S′) is the safety reward, Stab(S, A, S′) is the stability reward, Delay(S, A, S′) is the delay penalty, α, β, τ are adjustment factors used to balance the importance of each indicator; Step S5: Collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

2. A power supply station business indicator early warning method according to claim 1, characterized in that: The historical data related to the business indicators of the power supply station include the number of smart meters, the number of non-smart meters, the number of negative loss areas, the number of high loss areas, the power supply, the completion rate of line loss work orders, the success rate of line loss management, the number of repeated switching areas, the completion rate of collection work orders, the recovery rate of overdue electricity charges, the meter reading problem rate, the cumulative line loss rate, the cumulative line loss qualified rate, the average monthly collection success rate, the proportion of all collection areas, the online monitoring rate, the recovery rate of electricity charges in the previous month, the recovery rate of electricity charges in the current year, the amount of overdue charges in the current year, the amount of overdue charges in the current month, the ... The amount of electricity bills owed, the cumulative amount of arrears, the amount of liquidated damages for the year, the number of overdue work orders for residential business, the number of overdue work orders for non-residential business, the daily line loss qualification rate, the daily line loss rate, the daily data collection success rate, the number of users with daily data collection failures, the daily distribution transformer load qualification rate, the daily heavy load number, the daily overload number, the daily low voltage number, the electricity bill recovery rate for the current month, the number of users in arrears for the current month, the amount of arrears for the current month, the number of substations, the number of users, the number of low-voltage non-residential households, the number of low-voltage residential households, the number of high-voltage users, the number of substations, the number of lines, and the number of transformers.

3. A power supply station business indicator early warning method according to claim 1, characterized in that: The core of the intelligent agent based on the double-delay deep deterministic policy gradient algorithm is a double-delay neural network, which includes two target networks and one policy network.

4. A power supply station business indicator early warning system, characterized in that: It includes the following units: reinforcement learning environment construction unit, historical data collection unit, historical data screening unit, intelligent body construction unit, and early warning unit; The reinforcement learning environment construction unit is used to transform the real power system environment into a reinforcement learning environment capable of interacting with the intelligent agent; The historical data collection unit is used to collect historical data related to the business indicators of the power supply station and perform preprocessing; The historical data screening unit is used to screen the pre-processed historical data using an improved automatic encoding machine to obtain screened historical data; The training of the improved autoencoder is described as follows: h1=σ e (W1x+b1) (1) z=σ d (W2h1+b2) (2) Wherein, W1, b1 are the weight and bias of the encoding, W2, b2 are the weight and bias of the encoding, x is the input historical data, z is the encoding result, and the encoding result is the filtered historical data; h1 is the hidden layer variable, σ e , σ d are all constants; The loss function of the improved automatic encoder is described as follows: Among them, J(·) is the total loss function, L(·) is the sub-loss function; The intelligent agent construction unit is used to construct an intelligent agent based on a double-delayed deep deterministic policy gradient algorithm based on the filtered historical data; The core of the agent is a dual-delay neural network, which consists of two target networks Q1(S, A), Q2(S, A) and a policy network μ(s); The execution process of the double-delay neural network is as follows: For the current state S, S = z, perform action A: A=μ(s)+N (4) Where N is the random noise in the noise model; perform action A, get reward R and next state S′, store the experience (S, A, R, S′) in the experience buffer; randomly extract M experiences (S i , A i , R i , S′ i ), if S′ i is the terminal state, R i Assign to the value function; The training process of the double-delay neural network is as follows: The value function is the experience reward R i The agent first calculates the next action by passing the next state S′ in the sampled experience to the target Actor and adds noise to the action. The agent passes the next action to the target Critic to obtain the cumulative reward. In each training step, the Critic parameters are updated by minimizing the loss L of all sampled experiences: Among them, L k is the calculated loss, y i is the expected decision, Q k is the reinforcement learning weight factor, is the learnable parameter involved, i represents the i-th experience; The agent’s reward function is expressed as: R(S,A,S′)=α·Safe(S,A,S′)+β·Stab(S,A,S′)-τ·Delay(S,A,S′) (6) Among them, S represents the current state, A represents the agent action, S′ represents the next state, Safe(S, A, S′) is the safety reward, Stab(S, A, S′) is the stability reward, Delay(S, A, S′) is the delay penalty, α, β, τ are adjustment factors used to balance the importance of each indicator; The early warning unit is used to collect data related to the business indicators of the power supply station in real time, use the intelligent agent based on the double-delay deep deterministic policy gradient algorithm to monitor the data related to the business indicators of the power supply station collected in real time, and issue an early warning according to the early warning strategy.

5. A power supply station business indicator early warning system according to claim 4, characterized in that: The historical data related to the business indicators of the power supply station include the number of smart meters, the number of non-smart meters, the number of negative loss areas, the number of high loss areas, the power supply, the completion rate of line loss work orders, the success rate of line loss management, the number of repeated switching areas, the completion rate of collection work orders, the recovery rate of overdue electricity charges, the meter reading problem rate, the cumulative line loss rate, the cumulative line loss qualified rate, the average monthly collection success rate, the proportion of all collection areas, the online monitoring rate, the recovery rate of electricity charges in the previous month, the recovery rate of electricity charges in the current year, the amount of overdue charges in the current year, the amount of overdue charges in the current month, the ... The amount of electricity bills owed, the cumulative amount of arrears, the amount of liquidated damages for the year, the number of overdue work orders for residential business, the number of overdue work orders for non-residential business, the daily line loss qualification rate, the daily line loss rate, the daily data collection success rate, the number of users with daily data collection failures, the daily distribution transformer load qualification rate, the daily heavy load number, the daily overload number, the daily low voltage number, the electricity bill recovery rate for the current month, the number of users in arrears for the current month, the amount of arrears for the current month, the number of substations, the number of users, the number of low-voltage non-residential households, the number of low-voltage residential households, the number of high-voltage users, the number of substations, the number of lines, and the number of transformers.

6. A power supply station business indicator early warning system according to claim 4, characterized in that: The core of the intelligent agent based on the double-delay deep deterministic policy gradient algorithm is a double-delay neural network, which includes two target networks and one policy network.

7. A storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the power supply station business indicator early warning method as described in any one of claims 1-3.

8. An electronic terminal, characterized in that: The electronic terminal includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the power supply station business indicator early warning method as described in any one of claims 1-3.