Power supply service business hosting method, system and device and storage medium

Abstract

The invention relates to a power supply service business hosting method, system and device, and a storage medium. The method comprises the following steps: collecting enterprise electrical load data, preprocessing the data, and constructing a multi-modal feature vector; constructing an enterprise power consumption scene classification model, and inputting the multi-modal feature vector into the enterprise power consumption scene classification model to identify the power consumption scene of the enterprise; generating a trusteeship contract according to the power consumption scene of the enterprise, and executing trusteeship service according to the trusteeship contract; constructing an energy scheduling optimization model, and optimizing energy scheduling of the trusteeship enterprise through the energy scheduling optimization model during execution of the trusteeship service; and constructing an enterprise credit evaluation model, calculating a credit score of the enterprise in each period according to the load stability, the demand response cooperation degree, the electric charge payment timeliness and the electric energy quality cooperation degree of the enterprise, and adjusting the hosting contract parameter of the next period according to the credit score.

Description

Technical Field

[0001] The present invention relates to a power supply service business hosting method, system, equipment and storage medium, belonging to the technical field of power supply services. Background Art

[0002] Enterprise electricity demand is large-scale, highly continuous, and complex and volatile. Traditional power supply service models are unable to meet the requirements of both high reliability and high flexibility. Existing energy service models typically allow customers to subscribe to service providers for power equipment and energy management without upfront capital investment. However, these models often focus on energy efficiency and cost, lacking intelligent optimization and flexibility mechanisms. For example, while some existing energy management systems can collect and display electricity usage data, they often use rule-based or simple statistical methods to identify electricity usage scenarios. These methods struggle to effectively handle the complex multimodal, high-dimensional, and time-series characteristics of load data, resulting in insufficient scenario recognition accuracy and an inability to provide an accurate basis for refined services. Regarding energy scheduling, traditional models rely on manual experience or simple optimization models such as linear programming. These models struggle to adapt to the rapidly changing price signals and supply and demand fluctuations in the electricity market, and fail to dynamically consider the flexible response capabilities of user-side self-contained power supplies and energy storage equipment. This leads to rigid scheduling strategies and low resource utilization. Furthermore, existing user credit evaluations are often limited to a single dimension, such as electricity payment history. They lack a comprehensive assessment of comprehensive electricity usage behaviors, including load stability, demand response compliance, and power quality compliance. Credit evaluation results are also not dynamically linked to power supply service contracts, resulting in an inadequate incentive mechanism and making it difficult to effectively guide users to optimize their electricity usage behaviors and jointly maintain grid stability. Modern corporate electricity consumption is not limited to traditional manufacturing. There is also a large demand in information technology, service industries, logistics and other fields. The electricity consumption patterns and load changes in these fields are more complex and dynamic. Traditional models are difficult to meet such diverse needs. There is an urgent need for more intelligent and accurate data analysis tools to achieve real-time monitoring and predictive analysis. Summary of the Invention

[0003] In order to solve the above problems in the prior art, the present invention proposes a method, system, device and storage medium for power supply service hosting.

[0004] The technical solutions of the present invention are as follows: In one aspect, the present invention provides a method for entrusting a power supply service business, comprising the following steps: Collect enterprise electricity load data and construct it into a multimodal feature vector after preprocessing; Build an enterprise electricity usage scenario classification model and input the multimodal feature vector into the enterprise electricity usage scenario classification model to identify the enterprise's electricity usage scenario; According to the power consumption scene of the enterprise, a hosting contract is generated, and the hosting service is executed according to the hosting contract; An energy scheduling optimization model is constructed, and the energy scheduling of the hosting enterprise is optimized through the energy scheduling optimization model during the execution of the hosting service. An enterprise credit evaluation model is constructed, and the credit score of each period of the enterprise is calculated according to the load stability, demand response cooperation degree, electricity payment timeliness and power quality cooperation degree, and the hosting contract parameters of the next period are adjusted according to the credit score.

[0005] Preferably, the enterprise power load data includes active power, reactive power, voltage, total harmonic distortion rate, power frequency and equipment operation data.

[0006] The construction of the multi-modal feature vector includes synchronizing, aligning and normalizing different types and sources of power load data, and then converting the original data into a vector representation that can effectively represent the enterprise power consumption behavior and characteristics through feature engineering. Specifically, the load modal feature vector, the power quality modal feature vector and the equipment operation modal feature vector are aligned and spliced in the time dimension to form a unified multi-modal data input. This ensures that the subsequent model can simultaneously utilize information from different physical quantities for comprehensive analysis.

[0007] Preferably, the enterprise power consumption scene classification model includes a time domain feature multi-scale adaptive convolution layer, a modal fusion adaptive gating layer and a scene classification decision layer. A multi-modal feature vector is constructed based on the enterprise power load data, which includes a load modal feature vector, a power quality modal feature vector and an equipment operation modal feature vector. The time domain feature multi-scale adaptive convolution layer includes a short-term feature extraction unit, a medium-term feature extraction unit and a long-term feature extraction unit. The multi-modal feature vector is input into the short-term feature extraction unit, the medium-term feature extraction unit and the long-term feature extraction unit respectively to extract the short-term, medium-term and long-term features of the multi-modal feature vector, and then the multi-scale convolution feature vector is obtained by splicing, which is specifically shown in the following formula: ; Wherein: represents the multi-scale convolution feature vector at time t; represents the multi-modal feature vector at time t; represents the short-term feature of the multi-modal feature vector at time t; represents the medium-term feature of the multi-modal feature vector at time t; represents the long-term feature of the multi-modal feature vector at time t; represents the long-term feature of the multi-modal feature vector at time t; represents the long-term feature of the multi-modal feature vector at time t; represents the long-term feature of the multi-modal feature vector at time t; represents the long-term feature of the multi-modal feature vector at time t; Long-term features of the multimodal feature vector at each moment; Represents a splicing operation; The modality fusion adaptive gating layer includes an intramodal gating unit, a cross-modal interaction gating unit, and a feature interaction fusion layer. The intramodal gating factors of different modalities are extracted through the intramodal gating unit, as shown in the following formula: ; in: express time The intramodal gating factor of each modality; express The first multimodal feature vector at the moment The eigenvectors of the modes; Indicates the The intramodal weight matrix under each mode; Indicates the Intramodal bias under each mode; represents the activation function; The multimodal feature vector is input into the cross-modal interaction gating unit to extract the cross-modal interaction factors of different modalities, as shown in the following formula: ; in: express Moment Cross-modal interaction factors of the modalities; represents the total number of modes; Indicates the The modal pair The influence coefficient of each mode; express The first multimodal feature vector at the moment The eigenvectors of the modes; express The first multimodal feature vector at the moment The hidden state of each modality; Indicates the The hidden state weight matrix of each modality; Indicates the Cross-modal bias term under different modalities; The adaptive gating factor of each modality is obtained by fusing the intramodal gating factor of each modality with the cross-modal interaction factor through the feature interaction fusion layer, as shown in the following formula: ; in: express Moment Adaptive gating factors for each modality; The multi-scale convolution feature vector is weighted by the adaptive gating factor of each modality to obtain the hidden state of each modality, as shown in the following formula: ; in: express The first multimodal feature vector at the moment The hidden state of each modality; Normalize and merge the hidden states of each modality to obtain the modality fusion feature vector, as shown in the following formula: ; in: express The modal fusion feature vector at the moment; Represents normalization operation; The scene classification decision layer is constructed based on a multi-layer perceptron, as shown in the following formula: ; ; in: Indicates that the first layer of the multilayer perceptron is Output results at the moment; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias of the first layer of the multilayer perceptron; express function; Indicates that the second layer of the multilayer perceptron is The output result at the moment is the probability of the current enterprise in different electricity usage scenarios; Represents the weight matrix of the second layer of the multilayer perceptron; Represents the bias of the second layer of the multilayer perceptron.

[0008] Preferably, the hosting contract includes the enterprise's load capacity and electricity price, wherein the load capacity includes guaranteed basic load capacity and dynamic elastic load quota capacity; Allocate corresponding guaranteed basic load capacity to the enterprise based on its electricity usage scenario; The calculation formula for the dynamic elastic load quota capacity is: ; in: express Dynamic elastic load quota capacity at the moment; express Dynamic elastic load quota capacity at the moment; represents the motivation factor; Represents the enterprise Credit score at the moment; represents the penalty factor; Indicates that the enterprise The usage rate exceeds the dynamic elastic load quota capacity at all times; The electricity price is adaptively calculated based on the enterprise's electricity usage scenario, as shown in the following formula: ; in: Indicates that the enterprise The electricity price at the time; express Basic electricity price at the time; express The time period impact factor of the moment; express The elastic load utilization factor of the enterprise at the moment; Represents the market tension factor.

[0009] Preferably, the energy scheduling optimization model constructs an immediate reward function ,in, express The power consumption status at all times, express Electricity usage at all times; The electricity usage status specifically includes the enterprise's load demand, the enterprise's hosting contract electricity price, the enterprise's self-generated power unit price, and the enterprise's own energy storage usage cost; The electricity consumption actions specifically include the enterprise's purchased electricity power, the enterprise's self-generated power, and the enterprise's own energy storage power; The instant reward function is specifically shown as follows: ; ; in: express The enterprise hosting contract electricity price at the moment; express The power purchased by the enterprise at the time; express The unit price of the enterprise's self-generated power generation at the moment; express The enterprise's self-generated power generation capacity at any time; express The cost of using the company's own energy storage at that moment; express The company's own energy storage capacity at any given time; represents the balance penalty coefficient; express The imbalance between electricity supply and demand at each moment; express The load demand of the enterprise at all times; Construct an objective function that maximizes the immediate reward function, as shown in the following formula: ; in: Indicates the power action strategy; Indicates the total length of the decision time domain; express Discount factor for the moment; represents the mathematical expectation; The above objective function is solved based on the deep Q network. The specific steps are: The state-action reward function is defined based on the immediate reward function, as shown in the following formula: ; in: Indicates that the power is on Take electrical action expected rewards; Indicates that the power is on Take electrical action Then transfer to power state Calculate the expectation of the probability distribution; Indicates that the power is on Take the best electricity use action the maximum expected reward achieved; The parameters of the deep Q network are updated according to the following loss function: ; in: represents the loss function; Represents the current parameters of the deep Q network; Represents the experience replay buffer, which is used to store data samples from the previous iteration; Indicates that the power consumption status in the previous iteration is collected from the experience replay buffer , electrical action , instant rewards And power status Perform expected value calculations; represents the target parameters of the deep Q network; Indicates that the power is on Take the best electricity use action And the deep Q network target parameters are The maximum expected reward obtained when Indicates that the deep Q network is based on the current parameters In power state Next electricity action The expected reward after making a prediction.

[0010] Preferably, the enterprise credit evaluation model is specifically shown as follows: ; in: Represents the enterprise Credit score at the moment; Represents the enterprise Load stability at all times; Represents the enterprise The degree of coordination in responding to demand at any given moment; Represents the enterprise Timely payment of electricity bills at all times; Represents the enterprise Power quality coordination at all times.

[0011] On the other hand, the present invention also provides a power supply service business hosting system, including a data acquisition module, an enterprise power consumption scenario classification module, a hosting contract generation module, an enterprise energy scheduling optimization module, and an enterprise credit evaluation module; The data acquisition module is used to collect enterprise electricity load data and construct it into a multimodal feature vector after preprocessing; The enterprise electricity usage scenario classification module is used to build an enterprise electricity usage scenario classification model, and input the multimodal feature vector into the enterprise electricity usage scenario classification model to identify the enterprise's electricity usage scenario; The escrow contract generation module is used to generate an escrow contract based on the enterprise's electricity usage scenario and perform escrow services according to the escrow contract; The enterprise energy scheduling optimization module is used to build an energy scheduling optimization model, and optimize the energy scheduling of the managed enterprise through the energy scheduling optimization model during the execution of the managed service; The enterprise credit evaluation module is used to build an enterprise credit evaluation model, calculate the enterprise's credit score for each cycle based on the enterprise's load stability, demand response coordination, electricity bill payment timeliness, and power quality coordination, and adjust the custody contract parameters for the next cycle based on the credit score.

[0012] On the other hand, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present invention when executing the program.

[0013] In another aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any embodiment of the present invention.

[0014] The present invention has the following beneficial effects: 1. The present invention uses a multi-scale adaptive convolution layer with time domain features to extract short-term, medium-term and long-term features from enterprise load data, and then combines it with a modal fusion adaptive gating layer to deeply fuse multimodal data, thereby improving the recognition accuracy of power consumption scenarios.

[0015] 2. The present invention generates a hosting contract for the specific electricity usage scenarios of enterprises, achieving more flexible and efficient electricity price regulation. This dynamic adjustment mechanism is significantly superior to the traditional static pricing method and is conducive to improving the efficiency of power supply resource allocation.

[0016] 3. By constructing an energy scheduling optimization model, the present invention can achieve optimal regulation of supply and demand balance while taking into account multiple factors such as the company's electricity consumption status, electricity purchase, self-generated power generation and energy storage, overcoming the shortcomings of traditional methods that rely solely on empirical decision-making or simple models to solve scheduling problems.

[0017] 4. This invention constructs an enterprise credit evaluation model, calculates credit scores in real time, and dynamically adjusts the parameters of the next cycle's custody contract based on the scores. This mechanism not only helps to effectively control risks, but also incentivizes enterprises to improve their energy-saving performance, far exceeding the existing credit evaluation model based on a single performance record. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] It should be understood that the step numbers used herein are only for convenience of description and are not intended to limit the order in which the steps are to be executed.

[0021] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0023] The term "and / or" refers to and includes any and all possible combinations of one or more of the associated listed items.

[0024] Example 1: See also Figure 1 A method for entrusting a power supply service business comprises the following steps: Collect enterprise electricity load data and construct it into a multimodal feature vector after preprocessing; Build an enterprise electricity usage scenario classification model and input the multimodal feature vector into the enterprise electricity usage scenario classification model to identify the enterprise's electricity usage scenario; Generate a hosting contract based on the company's electricity usage scenario and perform hosting services according to the hosting contract; Construct an energy scheduling optimization model and optimize the energy scheduling of the hosting enterprise through the energy scheduling optimization model during the execution of the hosting service; Build an enterprise credit evaluation model to calculate the enterprise's credit score for each cycle based on its load stability, demand response coordination, electricity bill payment timeliness, and power quality coordination, and adjust the custody contract parameters for the next cycle based on the credit score.

[0025] The data collection methods are diversified, covering contact and non-contact collection methods, and combined with Robotic Process Automation (RPA) technology to improve efficiency and accuracy.

[0026] Specifically, contact-based IoT sensor collection methods include: personnel information collection equipment, environmental monitoring instruments and meters, video equipment, stress and strain sensors, hydropower instruments and meters, and other IoT collection devices. These transmit data to the system via a standard Ethernet interface or 4G / 5G network on a front-end communication management machine. This data includes active power, reactive power, voltage, total harmonic distortion, power frequency, as well as the device's operating temperature and operating current.

[0027] The collection methods of contactless smart wearable sensors include: collecting on-site process data and business collaboration data through contactless mobile devices such as mobile inspection equipment and smart wearable devices, and transmitting data through mobile transmission technologies such as 4G / 5G and Bluetooth.

[0028] The application of RPA technology in data collection is primarily reflected in: automatically connecting user-side terminal systems (such as intranet data platforms and smart meter terminals) with device sensor networks, extracting real-time data based on preset fields or by accessing sensor data gateways; automatically querying historical databases and batch exporting records for specified time periods; enabling real-time collection of high-frequency data at the millisecond / second level, as well as scheduled collection of periodic data at the minute / hour level; and automatically initiating targeted data collection when abnormal power usage is detected. After RPA collects data, it can directly connect to the data preprocessing process, automatically completing format standardization, preliminary outlier filtering, and data splicing.

[0029] As a preferred implementation of this embodiment, the enterprise power load data includes active power, reactive power, voltage, total harmonic distortion, power frequency and equipment operation data (specifically, operating temperature and operating current of the equipment).

[0030] In this embodiment, a steel smelting enterprise is taken as an example to collect 24-hour electricity load, as shown in the following table:

[0031] The above example data is collected at hourly intervals. The actual collection frequency can be set according to specific needs. For example, for scenarios with high real-time requirements, such as production power consumption monitoring of large industrial enterprises, millisecond-level collection can be used; for general commercial users, minute-level or hour-level collection can be used.

[0032] As a preferred implementation of this embodiment, the enterprise electricity consumption scenario classification model includes a time domain feature multi-scale adaptive convolution layer, a modality fusion adaptive gating layer, and a scenario classification decision layer; Constructing a multimodal feature vector based on the enterprise's electricity load data. The multimodal feature vector includes a load modal feature vector (active power, reactive power), a power quality modal feature vector (voltage, total harmonic distortion, power frequency), and an equipment operation modal feature vector (equipment operating temperature and operating current). The time domain feature multi-scale adaptive convolution layer includes a short-term feature extraction unit (convolution kernel size is 3), a medium-term feature extraction unit (convolution kernel size is 7), and a long-term feature extraction unit (convolution kernel size is 14). The multimodal feature vector is input into the short-term feature extraction unit, the medium-term feature extraction unit, and the long-term feature extraction unit respectively to extract the short-term, medium-term, and long-term features of the multimodal feature vector, and then splice them to obtain a multi-scale convolution feature vector, as shown in the following formula: ; in: express Multi-scale convolution feature vector at the moment; express Multimodal feature vector at time t; express Short-term features of the multimodal feature vector at the moment; express mid-term features of the multimodal feature vector at the moment; express Long-term features of the multimodal feature vector at each moment; Represents a splicing operation; The modality fusion adaptive gating layer includes an intramodal gating unit, a cross-modal interaction gating unit, and a feature interaction fusion layer. The intramodal gating factors of different modalities are extracted through the intramodal gating unit, as shown in the following formula: ; in: express time The intramodal gating factor of each modality; express The first multimodal feature vector at the moment The eigenvectors of the modes; Indicates the The intramodal weight matrix under each mode; Indicates the Intramodal bias under each mode; represents the activation function; The multimodal feature vector is input into the cross-modal interaction gating unit to extract the cross-modal interaction factors of different modalities, as shown in the following formula: ; in: express Moment Cross-modal interaction factors of the modalities; represents the total number of modes; Indicates the The modal pair The influence coefficient of each mode; express The first multimodal feature vector at the moment The eigenvector of each mode; express The first multimodal feature vector at the moment The hidden state of each modality; Indicates the The hidden state weight matrix of each modality; Indicates the Cross-modal bias term under different modalities; The adaptive gating factor of each modality is obtained by fusing the intramodal gating factor of each modality with the cross-modal interaction factor through the feature interaction fusion layer, as shown in the following formula: ; in: express Moment Adaptive gating factors for each modality; The multi-scale convolution feature vector is weighted by the adaptive gating factor of each modality to obtain the hidden state of each modality, as shown in the following formula: ; in: express The first multimodal feature vector at the moment The hidden state of each modality; Normalize and merge the hidden states of each modality to obtain the modality fusion feature vector, as shown in the following formula: ; in: express The modal fusion feature vector at the moment; Represents normalization operation; The scene classification decision layer is constructed based on a multi-layer perceptron, as shown in the following formula: ; ; in: Indicates that the first layer of the multilayer perceptron is Output results at the moment; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias of the first layer of the multilayer perceptron; express function; Indicates that the second layer of the multilayer perceptron is The output result at the moment is the probability of the current enterprise in different electricity usage scenarios; Represents the weight matrix of the second layer of the multilayer perceptron; Represents the bias of the second layer of the multilayer perceptron; The classification output is: Production continuity (continuous / discontinuous); Load fluctuation (severe / smooth); Power quality sensitivity (high / medium / low); These classification outputs collectively form a comprehensive profile of an enterprise's electricity usage scenarios. For example, "continuous production + severe fluctuations + high sensitivity to power quality" represents an industrial enterprise with extremely high requirements for power supply reliability and quality, and with rapidly fluctuating loads. These precise scenario identification results serve as the foundation for the subsequent generation of customized escrow contracts, refined energy scheduling, and risk management.

[0033] As a preferred implementation of this embodiment, the hosting contract includes the enterprise's load capacity and electricity price, wherein the load capacity includes guaranteed basic load capacity and dynamic elastic load quota capacity; The corresponding guaranteed basic load capacity is allocated to each enterprise based on its electricity usage scenario, as shown in the following table:

[0034] The calculation formula for the dynamic elastic load quota capacity is: ; in: express Dynamic elastic load quota capacity at the moment; express Dynamic elastic load quota capacity at the moment; represents the motivation factor; Indicates the enterprise Credit score at the moment; represents the penalty factor; Indicates that the enterprise The usage rate exceeds the dynamic elastic load quota capacity at all times; The electricity price is adaptively calculated based on the enterprise's electricity usage scenario, as shown in the following formula: ; in: Indicates that the enterprise The electricity price at the moment is obtained based on the long-term market electricity price forecast; express Basic electricity price at the time; express The time period impact factor is set based on the grid load at different time periods; express The elastic load utilization factor of the enterprise at the moment; represents the market tension factor; The elastic load utilization factor calculation formula is specifically as follows: ; in: express Flexible loads that are used by enterprises in real time; Indicates the dynamic elastic load quota capacity currently owned by the enterprise; represents the elastic load ladder price penalty coefficient; The market tension factor calculation formula is: ; in: represents the market sensitivity adjustment coefficient; express Total electricity demand in the electricity market at any given moment; express The total electricity supply in the electricity market at that moment.

[0035] As a preferred implementation of this embodiment, the energy scheduling optimization model constructs an instant reward function ,in, express The power consumption status at all times, express Electricity usage at all times; The electricity usage status specifically includes the enterprise's load demand, the enterprise's hosting contract electricity price, the enterprise's self-generated power unit price, and the enterprise's own energy storage usage cost; The electricity consumption actions specifically include the enterprise's purchased electricity power, the enterprise's self-generated power, and the enterprise's own energy storage power; The instant reward function is specifically shown as follows: ; ; in: express The enterprise hosting contract electricity price at the moment; express The power purchased by the enterprise at the time; express The unit price of the enterprise's self-generated power generation at the moment; express The enterprise's self-generated power generation capacity at any time; express The cost of using the company's own energy storage at that moment; express The company's own energy storage capacity at any given time; represents the balance penalty coefficient; express The imbalance between electricity supply and demand at each moment; express The load demand of the enterprise at all times; Construct an objective function that maximizes the immediate reward function, as shown in the following formula: ; in: Indicates the power action strategy; Indicates the total length of the decision time domain; express Discount factor for the moment; represents the mathematical expectation; The above objective function is solved based on the deep Q network. The specific steps are: The state-action reward function is defined based on the immediate reward function, as shown in the following formula: ; in: Indicates that the power is on Take electrical action expected rewards; Indicates that the power is on Take electrical action Then transfer to power state Calculate the expectation of the probability distribution; Indicates that the power is on Take the best electricity use action the maximum expected reward achieved; The parameters of the deep Q network are updated according to the following loss function: ; in: represents the loss function; Represents the current parameters of the deep Q network; Represents the experience replay buffer, which is used to store data samples from the previous iteration; Indicates that the power consumption status in the previous iteration is collected from the experience replay buffer , electrical action , instant rewards And power status Perform expected value calculations; represents the target parameters of the deep Q network; Indicates that the power is on Take the best electricity use action And the deep Q network target parameters are The maximum expected reward obtained when Indicates that the deep Q network is based on the current parameters In power state Next electricity action The expected reward after making a prediction.

[0036] As a preferred implementation of this embodiment, the enterprise credit evaluation model is specifically shown as follows: ; wherein: represents the credit score of the enterprise at the moment ; represents the load stability of the enterprise at the moment ; represents the demand response cooperation degree of the enterprise at the moment ; represents the electricity payment timeliness of the enterprise at the moment ; represents the power quality cooperation degree of the enterprise at the moment ; The calculation formula of the load stability is: ; wherein: represents the normalized value of the standard deviation of the power load of the enterprise to the moment ; represents the proportion of the peak load to the average load of the enterprise to the moment ; represents the power load of the enterprise at the moment ; represents the power load change rate of the enterprise at the moment ; The calculation formula of the demand response cooperation degree is: ; wherein: represents the total number of demand response event requests; represents the actual response proportion (0-1) of the enterprise to the th demand response event request at the moment ; represents the load reduction amount of the th demand response event request at the moment ; represents a natural constant; represents the response delay time of the enterprise to the th demand response event request at the moment ; represents the load response accuracy of the enterprise to the th demand response event request at the moment , that is, the absolute value proportion of the difference between the actual reduction amount and the required reduction amount; A demand response event is a temporary notification or request issued by a power supplier to a user (usually a large industrial or commercial user) under specific conditions (such as peak grid load, energy shortage, grid accident, or significant market price fluctuations). The request requires the user to proactively adjust their electricity usage within a specified timeframe to alleviate grid burden, reduce peak load, or stabilize grid operation. The calculation formula for the timeliness of electricity bill payment is: ; in: express The average delay in paying electricity bills for enterprises at the moment; express The frequency of defaults by the enterprise at any given moment; express The amount of the enterprise that is overdue and unpaid at the moment; express The total electricity bill payable by the enterprise at that moment; The calculation formula of the power quality coordination degree is: ; in: express The difference between the actual voltage and the rated voltage of the enterprise's electricity consumption at the moment; Indicates the rated voltage value of the enterprise; express Total harmonic distortion rate of the enterprise's load current at any moment; express The voltage flicker index of the enterprise at the moment is used to measure short-term voltage fluctuations and is set based on experimental parameters.

[0037] Example 2: A power supply service hosting system, applied to intelligent agents with perception and computing capabilities; The embodied intelligent agent, equipped with a built-in high-frequency data acquisition and processing module, can acquire and pre-process enterprise electricity load data within milliseconds and update multimodal feature vectors in real time. This advantage enables the system to instantly capture dynamic changes in enterprise electricity consumption and rapidly respond to emergencies and demand fluctuations, thereby achieving efficient energy scheduling and risk control. This millisecond-level responsiveness is crucial for power systems. For example, when responding to sudden load fluctuations, sudden grid failures (such as short circuits and voltage sags), or rapidly changing electricity market price signals, the embodied intelligent agent can provide faster decision-making and response times than traditional cloud computing or distributed systems. This effectively avoids potential losses caused by response delays, such as equipment damage, production interruptions, and power outages, and allows the system to seize market opportunities, achieving more refined and real-time supply and demand balancing and maximizing economic benefits. The embodied intelligent body integrates an enterprise electricity usage scenario classification module. This module uses a time-domain multi-scale adaptive convolutional layer to extract pre-processed multimodal feature vectors through short-term, medium-term, and long-term feature extraction. At the same time, it combines a modal fusion adaptive gating layer to achieve adaptive weighting of intramodal and cross-modal information, and finally outputs the current enterprise electricity usage scenario prediction result through the classification decision layer. Based on the electricity usage scenarios identified by the embodied intelligent agent, the contract generation module automatically generates an escrow contract. During the escrow service cycle, the contract generation module automatically adjusts the escrow contract parameters for the next cycle based on credit evaluation and energy scheduling feedback. During the hosting service, the embodied intelligent body deploys the energy scheduling optimization module and uses the instant reward function constructed by the deep Q network to perform real-time optimization scheduling of the company's purchased electricity power, self-generated power and self-owned energy storage power.

[0038] The embodied intelligent body has a built-in enterprise credit evaluation module, which calculates the enterprise's credit score in real time based on parameters such as the enterprise's load stability, demand response coordination, electricity bill payment timeliness, and power quality coordination.

[0039] The adaptive learning module of the embodied intelligent body updates the internal model parameters (including the convolution and gating parameters of the scene classification model, the weights in the deep Q network, etc.) based on the real-time scheduling results and feedback data output by the above modules, realizing closed-loop adaptive learning and optimization of the system, and improving future prediction accuracy and scheduling decision-making capabilities.

[0040] In this embodiment, the embodied intelligent body is equipped with a high-speed processor and large-capacity storage, ensuring that the device performs well in handling complex tasks. Equipped with a high-performance processor, memory, etc., it provides the device with powerful computing and storage capabilities, meeting the high requirements of the power industry for data processing and business applications; The embodied intelligent body provides multiple interfaces, including PSTN, Ethernet, USB, and display interfaces, to meet the connection needs of different devices. Four RJ11 PSTN interfaces support telephone system access, two RJ45 Ethernet ports support network connection, USB3.2 and USB2.0 interfaces facilitate external device connection, and HDMI and DP display interfaces support high-definition video output. The embodied intelligent body supports multiple communication protocols, including TCP / UDP, HTTP, HTTPS, ARP / RARP, DNS, NTP, TFTP, TELNET, STUN, etc., to achieve seamless connection and interconnection between different communication systems; it supports static IP address modification and DHCP dynamic allocation to meet the configuration requirements in different network environments; The embodied agent supports multiple audio codec formats such as G711A, G711U, G729, G723, iLBC, AMR-NB, etc., realizes multi-path concurrent calls, meets the demand of the power industry for efficient voice communication, supports mutual conversion between TDM calls and IP calls, and adapts to different communication environments and demands. The embodied agent is built-in with a large language model, can realize intelligent interactive response, supports complex dialogue scenarios and semantic understanding, the device can automatically identify the needs of users, provide accurate response and service, and improve user experience.

[0041] After adopting the agent technology, the whole power supply service business hosting process is highly automated: Data acquisition, scene recognition, contract generation, energy scheduling optimization and credit evaluation are all automatically executed by the agent, reducing the degree of human intervention, improving operation efficiency, and reducing operation and maintenance cost.

[0042] Embodiment three: The embodiment provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the program to realize the method in any embodiment of the application.

[0043] Embodiment four: The embodiment provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the method in any embodiment of the application.

[0044] In order to ensure the security and privacy of enterprise electricity data, the application adopts strict data security strategy in the whole process of data acquisition, processing, storage and application. For data that does not need to access real personal information, desensitization or anonymization is performed to reduce the risk of sensitive information leakage. At the same time, combined with the demand of application scene, safe multi-party calculation, federal learning and other technologies can be used to formulate cross-platform data joint calculation strategy, realize model training and data analysis without sharing original data, so as to realize the maximum utilization of data while ensuring the privacy of data.

[0045] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c or a and b and c, where a, b, c can be single or multiple.

[0046] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0047] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0048] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0049] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention's description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for entrusting power supply services, characterized in that: The following steps are involved: Collect enterprise electricity load data and construct it into a multimodal feature vector after preprocessing; Build an enterprise electricity usage scenario classification model and input the multimodal feature vector into the enterprise electricity usage scenario classification model to identify the enterprise's electricity usage scenario; Generate a hosting contract based on the company's electricity usage scenario and perform hosting services according to the hosting contract; Construct an energy scheduling optimization model and optimize the energy scheduling of the hosting enterprise through the energy scheduling optimization model during the execution of the hosting service; Build an enterprise credit evaluation model to calculate the enterprise's credit score for each cycle based on its load stability, demand response coordination, electricity bill payment timeliness, and power quality coordination, and adjust the custody contract parameters for the next cycle based on the credit score.

2. A power supply service business hosting method according to claim 1, characterized in that: The enterprise power load data includes active power, reactive power, voltage, total harmonic distortion, power frequency and equipment operation data.

3. The power supply service business hosting method according to claim 1, characterized in that: The enterprise electricity consumption scenario classification model includes a time domain feature multi-scale adaptive convolution layer, a modal fusion adaptive gating layer, and a scenario classification decision layer; Constructing a multimodal feature vector based on the enterprise's electricity load data, wherein the multimodal feature vector includes a load modal feature vector, a power quality modal feature vector, and an equipment operation modal feature vector; The time domain feature multi-scale adaptive convolution layer includes a short-term feature extraction unit, a medium-term feature extraction unit, and a long-term feature extraction unit. The multimodal feature vector is input into the short-term feature extraction unit, the medium-term feature extraction unit, and the long-term feature extraction unit respectively to extract the short-term, medium-term, and long-term features of the multimodal feature vector, and then spliced ​​to obtain a multi-scale convolution feature vector, as shown in the following formula: ； in: express Multi-scale convolution feature vector at the moment; express Multimodal feature vector at time t; express Short-term features of the multimodal feature vector at the moment; express mid-term features of the multimodal feature vector at the moment; express Long-term features of the multimodal feature vector at each moment; Represents a splicing operation; The modality fusion adaptive gating layer includes an intramodal gating unit, a cross-modal interaction gating unit, and a feature interaction fusion layer. The intramodal gating factors of different modalities are extracted through the intramodal gating unit, as shown in the following formula: ； in: express time The intramodal gating factor of each modality; express The first multimodal feature vector at the moment The eigenvectors of the modes; Indicates the The intramodal weight matrix under each mode; Indicates the Intramodal bias under each mode; represents the activation function; The multimodal feature vector is input into the cross-modal interaction gating unit to extract the cross-modal interaction factors of different modalities, as shown in the following formula: ； in: express Moment Cross-modal interaction factors of the modalities; represents the total number of modes; Indicates the The modal pair The influence coefficient of each mode; express The first multimodal feature vector at the moment The eigenvectors of the modes; express The first multimodal feature vector at the moment The hidden state of each modality; Indicates the The hidden state weight matrix of each modality; Indicates the Cross-modal bias term under different modalities; The adaptive gating factor of each modality is obtained by fusing the intramodal gating factor of each modality with the cross-modal interaction factor through the feature interaction fusion layer, as shown in the following formula: ； in: express Moment Adaptive gating factors for each modality; The multi-scale convolution feature vector is weighted by the adaptive gating factor of each modality to obtain the hidden state of each modality, as shown in the following formula: ； in: express The first multimodal feature vector at the moment The hidden state of each modality; Normalize and merge the hidden states of each modality to obtain the modality fusion feature vector, as shown in the following formula: ； in: express The modal fusion feature vector at the moment; Represents normalization operation; The scene classification decision layer is constructed based on a multi-layer perceptron, as shown in the following formula: ； ； in: Indicates that the first layer of the multilayer perceptron is Output results at the moment; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias of the first layer of the multilayer perceptron; express function; Indicates that the second layer of the multilayer perceptron is The output result at the moment is the probability of the current enterprise in different electricity usage scenarios; Represents the weight matrix of the second layer of the multilayer perceptron; Represents the bias of the second layer of the multilayer perceptron.

4. The method for entrusting power supply services according to claim 1, characterized in that: The hosting contract includes the enterprise's load capacity and electricity price, wherein the load capacity includes guaranteed basic load capacity and dynamic elastic load quota capacity; Allocate corresponding guaranteed basic load capacity to the enterprise based on its electricity usage scenario; The calculation formula for the dynamic elastic load quota capacity is: ； in: express Dynamic elastic load quota capacity at the moment; express Dynamic elastic load quota capacity at the moment; represents the motivation factor; Represents the enterprise Credit score at the moment; represents the penalty factor; Indicates that the enterprise The usage rate exceeds the dynamic elastic load quota capacity at all times; The electricity price is adaptively calculated based on the enterprise's electricity usage scenario, as shown in the following formula: ； in: Indicates that the enterprise The electricity price at the time; express Basic electricity price at the time; express The time period impact factor of the moment; express The elastic load utilization factor of the enterprise at the moment; Represents the market tension factor.

5. The power supply service business hosting method according to claim 1, characterized in that: The energy scheduling optimization model constructs an immediate reward function ,in, express The power consumption status at all times, express Electricity usage at all times; The electricity usage status specifically includes the enterprise's load demand, the enterprise's hosting contract electricity price, the enterprise's self-generated power unit price, and the enterprise's own energy storage usage cost; The electricity consumption actions specifically include the enterprise's purchased electricity power, the enterprise's self-generated power, and the enterprise's own energy storage power; The instant reward function is specifically shown as follows: ； ； in: express The enterprise hosting contract electricity price at the moment; express The power purchased by the enterprise at the time; express The unit price of the enterprise's self-generated power generation at the moment; express The enterprise's self-generated power generation capacity at any time; express The cost of using the company's own energy storage at that moment; express The company's own energy storage capacity at any given time; represents the balance penalty coefficient; express The imbalance between electricity supply and demand at each moment; express The load demand of the enterprise at all times; Construct an objective function that maximizes the immediate reward function, as shown in the following formula: ； in: Indicates the power action strategy; Indicates the total length of the decision time domain; express Discount factor for the moment; represents the mathematical expectation; The above objective function is solved based on the deep Q network. The specific steps are: The state-action reward function is defined based on the immediate reward function, as shown in the following formula: ； in: Indicates that the power is on Take electrical action expected rewards; Indicates that the power is on Take electrical action Then transfer to power state Calculate the expectation of the probability distribution; Indicates that the power is on Take the best electricity use action the maximum expected reward achieved; The parameters of the deep Q network are updated according to the following loss function: ； in: represents the loss function; Represents the current parameters of the deep Q network; Represents the experience replay buffer, which is used to store data samples from the previous iteration; Indicates that the power consumption status in the previous iteration is collected from the experience replay buffer , electrical action , instant rewards And power status Perform expected value calculations; represents the target parameters of the deep Q network; Indicates that the power is on Take the best electricity use action And the deep Q network target parameters are The maximum expected reward obtained when Indicates that the deep Q network is based on the current parameters In power state Next electricity action The expected reward after making a prediction.

6. The method for entrusting power supply services according to claim 1, characterized in that: The enterprise credit evaluation model is specifically shown as follows: ； in: Represents the enterprise Credit score at the moment; Represents the enterprise Load stability at all times; Represents the enterprise The degree of coordination in responding to demand at any given moment; Represents the enterprise Timely payment of electricity bills at all times; Represents the enterprise Power quality coordination at all times; 、 、 、 Represents the weight parameter of the enterprise credit evaluation model.

7. A power supply service hosting system, characterized in that: It includes data collection module, enterprise electricity usage scenario classification module, custody contract generation module, enterprise energy scheduling optimization module and enterprise credit evaluation module; The data acquisition module is used to collect enterprise electricity load data and construct it into a multimodal feature vector after preprocessing; The enterprise electricity usage scenario classification module is used to build an enterprise electricity usage scenario classification model, and input the multimodal feature vector into the enterprise electricity usage scenario classification model to identify the enterprise's electricity usage scenario; The escrow contract generation module is used to generate an escrow contract based on the enterprise's electricity usage scenario and perform escrow services according to the escrow contract; The enterprise energy scheduling optimization module is used to build an energy scheduling optimization model, and optimize the energy scheduling of the managed enterprise through the energy scheduling optimization model during the execution of the managed service; The enterprise credit evaluation module is used to build an enterprise credit evaluation model, calculate the enterprise's credit score for each cycle based on the enterprise's load stability, demand response coordination, electricity bill payment timeliness, and power quality coordination, and adjust the custody contract parameters for the next cycle based on the credit score.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images