Power transaction optimization method and system for industrial and commercial energy storage

Abstract

The invention relates to the technical field of power transaction optimization, in particular to a power transaction optimization method and system oriented to industrial and commercial energy storage, and the method comprises the steps: collecting power transaction dynamic parameters in real time, and obtaining power transaction historical feature data; performing preprocessing to obtain a standardized power transaction dynamic parameter set and a standardized historical parameter set; inputting the standardized historical parameter set into a pre-trained power transaction index prediction model, generating a power transaction prediction index, and making a power transaction plan based on the power transaction prediction index; converting the electric power transaction plan into a real-time constraint condition, inputting the electric power transaction prediction index and the standardized electric power transaction dynamic parameter set into a pre-trained transaction strategy generation model, and generating a charging and discharging power instruction meeting the real-time constraint condition; and the energy storage operator executes the real-time charging and discharging power instruction. According to the method, a closed-loop optimization method from data fusion and risk quantification to multi-objective decision is constructed, and collaborative optimization of multiple types of benefits can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transaction optimization, in particular to a power transaction optimization method and system for industrial and commercial energy storage. BACKGROUND

[0002] With large-scale access of renewable energy to the power system and continuous growth of industrial and commercial electricity demand, power supply and demand balance and system stability face severe challenges. As an important carrier of regulating power resources, industrial and commercial energy storage plays a key role in power transactions, and its efficient operation is of great significance to reduce electricity costs and improve new energy consumption capacity.

[0003] The prior art usually uses a multi-objective optimization model to coordinate the charging and discharging behavior of energy storage, for example, generates a transaction strategy through a joint optimization framework of electricity price prediction and load prediction, and designs multiple revenue modes such as peak-valley arbitrage and auxiliary services by combining a dynamic programming algorithm; in addition, some schemes use machine learning algorithms to predict electricity price fluctuation trends to provide decision basis for energy storage charging and discharging plans.

[0004] However, the existing coordination method of energy storage charging and discharging behavior has the following limitations: the transaction strategy relies on offline collection of single subject data (such as only considering grid electricity price or user load), and there are barriers to real-time interaction of multi-source dynamic parameters; the comprehensive quantification ability of electricity price peak-valley difference, load fluctuation and randomness of renewable energy output is insufficient, and it is difficult to generate a transaction plan that takes into account economic efficiency and robustness; the optimization goal is focused on a single peak-valley arbitrage mode, and multiple types of revenue such as capacity leasing revenue and green electricity transaction premium are not included in the unified decision framework, limiting the deep mining of energy storage value. SUMMARY

[0005] In view of the technical problems that the existing coordination method of energy storage charging and discharging behavior has insufficient multi-subject data cooperation, leading to decision-making relying on local information, the comprehensive quantification ability of electricity price fluctuation, load change and randomness of renewable energy output is lacking, and the optimization goal is single, limiting the deep mining of energy storage value, the present application provides a power transaction optimization method and system for industrial and commercial energy storage, which realizes multi-type revenue collaborative optimization through collaborative collection of power transaction dynamic parameters and historical characteristic data, multi-dimensional quantification indexes generated by prediction models and strategy generation under real-time constraints, and construction of a full-process closed loop from data fusion, risk quantification to multi-objective decision-making.

[0006] In the first aspect, the present application provides a power transaction optimization method for industrial and commercial energy storage, comprising the following steps: S1. Real-time collection of power transaction dynamic parameters and acquisition of power transaction historical characteristic data; S2. Preprocessing the power transaction dynamic parameters and transaction historical characteristic data to obtain a set of standardized power transaction dynamic parameters and a set of standardized historical parameters; S3. inputting the standardized historical parameter set into the pre-trained power transaction index prediction model to generate a power transaction prediction index in a specified time period, and formulating a power transaction plan for the same time period based on the power transaction prediction index; The power transaction prediction index includes a peak-valley price difference mean value, an industrial and commercial user electricity load fluctuation amplitude, and a renewable energy output confidence interval. S4. converting the power transaction plan into real-time constraint conditions, inputting the power transaction prediction index and the standardized power transaction dynamic parameter set into a pre-trained transaction strategy generation model to generate a charging and discharging power instruction that meets the real-time constraint conditions; S5. The energy storage operator executes the real-time charging and discharging power instruction.

[0007] Further, in step S1, the power transaction dynamic parameters include: The power grid price signal and the power grid operation data provided by the power grid company, the power grid operation data including the load, voltage and frequency of the power grid; The electricity load data and the energy storage system state of charge data provided by the industrial and commercial users, the energy storage system state of charge data including the charging and discharging power, temperature and voltage of the energy storage system; The energy storage device operation state parameters provided by the energy storage operator; The renewable energy output data provided by the renewable energy supplier; The power transaction historical feature data includes: The historical price record; The historical record of industrial and commercial user electricity load; The historical record of renewable energy output.

[0008] Further, in step S1, the power transaction dynamic parameters and the power transaction historical feature data are securely shared and credibly verified through blockchain or smart contract technology.

[0009] Further, step S2 includes: Data filtering, missing value filling and outlier correction are performed on the power transaction dynamic parameters to obtain a standardized power transaction dynamic parameter set; The missing value filling and normalization are performed on the transaction historical feature data to generate a standardized historical parameter set.

[0010] Further, in step S3, the power transaction index prediction model includes a time series feature extraction submodule, a spatial feature extraction submodule and a fusion output submodule, wherein: The time series feature extraction submodule is constructed based on a long short-term memory network and includes: The time sequence feature input layer receives time sequence data of historical electricity price records and historical load records of industrial and commercial users; The feature extraction LSTM network layer includes a feature extraction forget gate, a feature extraction input gate, and a feature extraction output gate. The feature extraction forget gate screens long-term trend features of historical electricity prices. The feature extraction input gate extracts short-term correlation features of load fluctuations of industrial and commercial users. The feature extraction output gate generates a time sequence feature hidden state vector; The fully connected compression layer includes a first branch and a second branch. The first branch reduces the dimension of the time sequence feature hidden state vector to output an electricity price fluctuation trend vector. The second branch reduces the dimension of the time sequence feature hidden state vector to output a load change trend vector. The spatial feature extraction submodule is based on a convolutional neural network and includes: The spatial feature input layer receives historical renewable energy output records and organizes them into a three-dimensional feature map with a width, a height, and a number of channels according to the meteorological stations where the renewable energy power generation equipment is located. The convolutional layer extracts regional meteorological correlation features by scanning the three-dimensional feature map with a 3x3 convolutional kernel and outputs a convolutional feature map. The activation layer performs ReLU nonlinear activation on the convolutional feature map and outputs an activated feature map. The max pooling layer performs 2x2 window maximum downsampling on the activated feature map and outputs a downsampled feature map. The flattening layer flattens the downsampled feature map into a renewable energy output feature vector. The fusion output submodule includes: The feature concatenation layer concatenates the electricity price fluctuation trend vector, the load change trend vector, and the renewable energy output feature vector into a fusion feature vector. The fully connected regression layer performs weighted calculation on the fusion feature vector and outputs the average peak-valley electricity price difference, the load fluctuation amplitude of industrial and commercial users, and the renewable energy output prediction value in a specified time period. The random disturbance branch layer adds Gaussian noise to the renewable energy output prediction value and then generates a renewable energy output confidence interval through Monte Carlo sampling.

[0011] Further, the random disturbance branch layer adds Gaussian noise with a mean of 0 and a standard deviation of the fluctuation rate of the historical renewable energy output records to the renewable energy output prediction value. Then, through 1000 times of Monte Carlo sampling, a renewable energy output confidence interval with a confidence level of 90% is generated.

[0012] Further, in step S3, formulating a power trading plan based on the power trading prediction index includes the following steps: S301. Construct a multi-objective optimization function:

[0013] wherein, is the time period of the power transaction index prediction model output is the average of the peak-valley price difference within the time period is the predicted total electricity load amount within the time period , which is equal to the average electricity load amount per unit time multiplied by the length of the time period, wherein the average electricity load amount per unit time is calculated based on the historical electricity load records of industrial and commercial users; is the capacity leasing revenue, , is the leased capacity, is the preset unit price of the capacity leasing; is the ancillary service revenue, , is the ancillary service proportion coefficient, is the preset unit price of the ancillary service; is the green electricity transaction amount; is the preset green electricity premium rate; , , is the weight coefficient, ; S302. Based on the new energy processing confidence interval, three output scenarios are set: a pessimistic scenario, the pessimistic scenario output , the optimistic probability weight is set to ; a benchmark scenario, the benchmark scenario output , the benchmark probability weight is set to ; an optimistic scenario, the optimistic scenario output , the optimistic probability weight is set to ; while satisfying: ; S303. The multi-objective optimization function is used as the objective function of the stochastic dual dynamic programming, and the decision variables are solved by the stochastic dual dynamic programming, including the power transaction amount , the leased capacity , and the green electricity transaction amount ; wherein, the solution of the power transaction amount is used to optimize the component, and the calculation formula is:

[0014] a fluctuation range of electricity consumption of the industrial and commercial user output by the electricity transaction index prediction model; a preset peak-valley adjustment coefficient; a new energy penetration rate, which is a ratio of a historical average processing of renewable energy to a historical average electricity consumption of the industrial and commercial user; a leasing capacity is solved for optimization component, and a calculation formula is:

[0015] a preset confidence correction factor; a green electricity transaction amount is solved for optimization component, and a calculation formula is:

[0016] S304. Output the electricity transaction plan, and the electricity transaction plan includes: a power contract scheme, including an electricity transaction amount , an electricity transaction price , , a preset price adjustment coefficient; a capacity leasing scheme, including a leasing capacity , a leasing period, and the leasing period is equal to a time period ; a green electricity transaction scheme, including a green electricity transaction amount , a green electricity premium rate .

[0017] It is further needed to be explained that, in step S4, the real-time constraint conditions include: a charging and discharging power upper limit constraint:

[0018]

[0019] wherein, is a charging power or a discharging power in a time period ; is an upper limit value of the charging and discharging power in the time period ; is an hour representation of a time period ; a preset charging and discharging efficiency coefficient; State of Charge reservation constraint:

[0020]

[0021] wherein, is State of Charge of the time period; is the minimum reserved State of Charge; is the rated capacity of the energy storage system; Daily green electricity trading volume upper limit constraint:

[0022]

[0023] is the green electricity trading volume of the day; is the daily green electricity trading volume upper limit; is the number of days in the time period .

[0024] Further, in step S4, the real-time constraint conditions further include: Energy storage physical constraints: the upper limit of charging and discharging power is 90% of the rated power of the energy storage device, and the State of Charge range is 20% to 90%; Grid operation constraints: the voltage fluctuation range does not exceed ±5% of the nominal voltage, and the frequency fluctuation range does not exceed ±0.2 Hz of the nominal frequency; Trading rule constraints: single transaction volume is 1 to 10 MWh, and transaction time window is 1 hour before the opening of the day-ahead market to the closing of the market.

[0025] Further, in step S4, the trading strategy generation model includes a medium and long-term planning sub-module and a real-time decision optimization sub-module, wherein: The medium and long-term planning sub-module is constructed using a long short-term memory network, including: Dynamic parameter input layer: receives power trading prediction indicators, and combines power trading prediction indicators of continuous N specified time periods into a multi-period prediction indicator group; Strategy generation long short-term memory network layer: including strategy generation forget gate, strategy generation input gate and strategy generation output gate, the strategy generation forget gate filters long-term trend features of the multi-period prediction indicator group, the strategy generation input gate extracts short-term fluctuation features of the multi-period prediction indicator group, and the strategy generation output gate generates a prediction indicator hidden state vector; Fully connected output layer: reduce the dimensionality of the prediction indicator hidden state vector, and output the medium and long term charging and discharging plan; The real-time decision optimization submodule is constructed using a proximal policy optimization algorithm, and includes: State input layer: receives the standardized power transaction dynamic parameter set, the medium and long term discharging plan and the real-time constraint condition, and then performs feature splicing to form a state vector; Policy network layer: receives the state vector and calculates the action probability distribution through a fully connected neural network; Value network layer: receives the state vector and calculates the state value through a fully connected neural network; Action output layer: receives the action probability distribution and the state value, selects the optimal action through the proximal policy optimization algorithm, and outputs the charging and discharging power instruction that meets the real-time constraint condition.

[0026] Further, the medium and long term charging and discharging plan output by the medium and long term planning submodule is a structured vector :

[0027] wherein: is the number of hours in the time period ; represents the target charging and discharging power baseline value of the future hour, when discharging, when charging.

[0028] Further, step S5 monitors and feeds back in real time during the execution of the charging and discharging power instruction, including: Real-time acquisition of power grid frequency data and energy storage device operating state; When it is detected that the power grid frequency fluctuation exceeds ±0.2Hz or the energy storage device temperature exceeds 60℃, an abnormal handling mechanism is triggered; The abnormal handling mechanism includes: suspending the current charging and discharging instruction, re-solving the transaction strategy generation model to generate an adjusted instruction, and issuing and executing within 10 seconds.

[0029] In a second aspect, the present application provides a power transaction optimization system for industrial and commercial energy storage, for implementing the above power transaction optimization method, including: A data acquisition module for real-time acquisition of power transaction dynamic parameters and obtaining power transaction historical feature data; A data preprocessing module for preprocessing the power transaction dynamic parameters and transaction historical feature data to obtain a standardized power transaction dynamic parameter set and a standardized historical parameter set; The index prediction and plan making module is configured to input the standardized historical parameter set into a pre-trained power transaction index prediction model to generate power transaction prediction indexes in a specified time period, and make a power transaction plan for the same time period based on the power transaction prediction indexes. The instruction generation module is configured to convert the power transaction plan into real-time constraint conditions, input the power transaction prediction indexes and the standardized power transaction dynamic parameter set into a pre-trained transaction strategy generation model, and generate charging and discharging power instructions that meet the real-time constraint conditions. The execution module is configured to execute the real-time charging and discharging power instructions.

[0030] In a third aspect, the present application provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor is configured to implement the steps of the power transaction optimization method when executing the computer program.

[0031] In a fourth aspect, the present application provides a storage medium, which stores a computer program, and the computer program is configured to implement the steps of the power transaction optimization method when executed by a processor.

[0032] From the above technical solutions, the present application has the following advantages: 1. The present application breaks through the multi-source data interaction barrier by collecting real-time power transaction dynamic parameters and power transaction historical characteristic data, realizes real-time collaborative collection of multi-agent dynamic parameters, and provides a complete data basis for strategy optimization.

[0033] 2. The present application generates a standardized data set based on preprocessing, cooperatively outputs the peak-valley price difference mean, industrial and commercial user electricity load fluctuation amplitude, and renewable energy output confidence interval through a power transaction index prediction model, and makes a power transaction plan accordingly, thereby improving the generation accuracy and cooperativeness of multi-dimensional key prediction indexes.

[0034] 3. The present application converts the power transaction plan into real-time constraint conditions, dynamically generates charging and discharging power instructions by a transaction strategy generation model in combination with the prediction indexes and the standardized dynamic parameter set, establishes a dynamic conversion mechanism from the plan to the instructions, and ensures the matching of the instructions and real-time working conditions.

[0035] 4. The present application monitors the power grid frequency and the state of the energy storage device in real time during the execution of the charging and discharging power instructions, automatically triggers instruction adjustment when detecting frequency overrun or temperature abnormality, forms a closed-loop feedback control of the execution link, and guarantees the safety of system operation.

[0036] 5. The application realizes the whole-cycle autonomous decision and collaborative optimization of industrial and commercial energy storage systems in the electricity market by constructing a complete technical chain including dynamic parameter acquisition, data standardization processing, electricity transaction index prediction, charge and discharge instruction generation and execution, ensuring that the energy storage operator can dynamically adjust the transaction strategy based on the prediction results of multi-source data fusion, maximize the comprehensive income of peak-valley arbitrage, capacity leasing and green electricity transaction under the condition of meeting the real-time constraints of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the present application, the drawings needed to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0038] Figure 1 is a flowchart of the electricity transaction optimization method for industrial and commercial energy storage in an embodiment of the present application.

[0039] Figure 2 is a schematic block diagram of the electricity transaction optimization system for industrial and commercial energy storage in an embodiment of the present application.

[0040] Figure 3 is a schematic diagram of the hardware structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the application purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions protected by the present application will be described in detail below with specific embodiments and drawings. Obviously, the following described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present patent, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present patent.

[0042] The electricity transaction optimization method related to the present application will be described in detail below. For the purpose of illustration rather than limitation, specific details such as specific system structures, techniques, etc. are presented in order to thoroughly understand the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details.

[0043] The term "comprising" is used in the sense of "including", and specifies the presence of stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The terms "comprising", "including", "containing", "have" and "including" and their conjugates mean "including, but not limited to", unless otherwise expressly stated.

[0044] In order to clearly describe the technical solutions of the present application, the terms "first", "second", etc. are used to distinguish the same or similar items with basically the same function and role. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the quantity and execution order, and the terms "first", "second", etc. also do not necessarily mean different.

[0045] The phrases "one embodiment", "an embodiment", "some embodiments", and the like, as used in the present application, mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Therefore, the occurrences of the phrases "in one embodiment", "in some embodiments", "in other embodiments", "in additional embodiments", and the like, in various places in the present application do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise expressly specified.

[0046] The technical solutions in the embodiments of the present application will be described clearly and completely in connection with the drawings in the embodiments of the present application.

[0047] The power transaction optimization method provided by the embodiments of the present application is executed by a computer device, and accordingly, the power transaction optimization system for industrial and commercial energy storage runs in the computer device.

[0048] Figure 1 is a flowchart of the power transaction optimization method for industrial and commercial energy storage according to one embodiment of the present application. In the flowchart, Figure 1 The execution subject can be a power transaction optimization system. According to different needs, the order of steps in the flowchart can be changed, and some steps can be omitted.

[0049] As Figure 1 indicated, the power transaction optimization method for industrial and commercial energy storage includes: Step S1, real-time collection of power transaction dynamic parameters and acquisition of power transaction historical feature data.

[0050] By real-time collection of power transaction dynamic parameters and acquisition of power transaction historical feature data, synchronous monitoring of key factors in the power market environment and extraction of historical behavior characteristics are realized, providing basic data input for subsequent standardized processing and prediction models.

[0051] In some embodiments, the power transaction dynamic parameters include: The power grid price signal provided by the power grid company, the power grid operation data including the load, voltage and frequency of the power grid; The power consumption load data provided by the industrial and commercial users, the state of charge data of the energy storage system, including the charge and discharge power, temperature and voltage of the energy storage system; The energy storage device operation state parameters provided by the energy storage operator; The renewable energy supply data provided by the renewable energy supplier; The power transaction historical feature data includes: Historical price records; Historical records of industrial and commercial user power consumption load; Renewable energy output historical records.

[0052] By defining the specific content of the power transaction dynamic parameters and historical feature data, the specific composition of the data collection range is clear.

[0053] In some embodiments, the power transaction dynamic parameters and the power transaction historical feature data are securely shared and credibly verified through blockchain or smart contract technology.

[0054] By using blockchain or smart contract technology to realize the secure sharing and credible verification of power transaction dynamic parameters and historical feature data, the tamper-proofing ability of the data exchange process is improved.

[0055] In some embodiments, the process of secure sharing and credible verification through blockchain technology includes: The subjects such as the power grid company and the industrial and commercial users generate unique hash values from the original data collected in real time through SHA-256 algorithm and upload them to the distributed ledger of Hyperledger Fabric consortium chain; The smart contract deployed on the chain automatically triggers the verifyData verification program to compare the consistency of the data hash to be verified with the stored value on the chain, and only when they are completely matched, the data is determined to be complete and credible; Through Chainlink oracle, the standardized parameter set that passes the verification is securely transmitted to the collaborative control module to start the power transaction prediction process; An abnormal fuse mechanism is established, if the same subject fails to verify for three consecutive times, its data upload permission will be automatically frozen and an alarm signal will be triggered, and the operation and maintenance personnel will intervene to check the equipment failure or malicious attack.

[0056] The above process is protected by cryptography, programmed verification and permission fuse, and builds an unalterable multi-subject collaborative data foundation.

[0057] Step S2, preprocessing the power transaction dynamic parameter and transaction history characteristic data to obtain a standardized power transaction dynamic parameter set and a standardized history parameter set.

[0058] By preprocessing the power transaction dynamic parameter and transaction history characteristic data to generate a standardized power transaction dynamic parameter set and a standardized history parameter set, the scale difference and non-normative of the original data are eliminated, and the compatibility of subsequent model processing is improved.

[0059] In some specific embodiments, the power transaction dynamic parameter is subjected to data filtering, missing value filling and abnormal value correction to obtain a standardized power transaction dynamic parameter set. The transaction history characteristic data is subjected to missing value filling and normalization to generate a standardized history parameter set.

[0060] By defining the specific technical means of standardization processing, the preprocessing operation is defined to include performing data filtering, missing value filling and abnormal value correction on the dynamic parameter to generate a standardized dynamic parameter set, and performing missing value filling and normalization on the history data to generate a standardized history parameter set.

[0061] Step S3, inputting the standardized history parameter set into a pre-trained power transaction index prediction model to generate power transaction prediction indexes within a specified time period, and formulating a power transaction plan for the same time period based on the power transaction prediction indexes. The power transaction prediction indexes include the average of peak-valley price difference, the fluctuation amplitude of industrial and commercial user electricity load, and the confidence interval of renewable energy output.

[0062] By inputting the standardized history parameter set into the pre-trained model to generate the average of peak-valley price difference, the fluctuation amplitude of electricity load, and the confidence interval of renewable energy output within a specified time period, and formulating a power transaction plan accordingly, a market decision basis based on quantitative prediction indexes is provided.

[0063] In some specific embodiments, the power transaction index prediction model includes a time series feature extraction submodule, a spatial feature extraction submodule, and a fusion output submodule, wherein: The time series feature extraction submodule is constructed based on a long short-term memory network and includes: A time series feature input layer: receiving time series data of historical electricity price records and industrial and commercial user electricity load history records; A feature extraction long short-term memory network layer: including a feature extraction forget gate, a feature extraction input gate, and a feature extraction output gate, the feature extraction forget gate filters long-term trend features of historical electricity prices, the feature extraction input gate extracts short-term correlation features of industrial and commercial user electricity load fluctuations, and the feature extraction output gate generates a time series feature hidden state vector; The full connection compression layer comprises a first branch and a second branch, the first branch reduces dimension of the time sequence feature hidden state vector to output the electricity price fluctuation trend vector, and the second branch reduces dimension of the time sequence feature hidden state vector to output the electricity load change trend vector; The spatial feature extraction submodule is constructed based on a convolutional neural network and comprises: The spatial feature input layer receives renewable energy output historical records and organizes the records into a three-dimensional feature map with a width, a height, and a number of channels according to the meteorological stations where the renewable energy power generation equipment is located. The convolutional layer extracts regional meteorological correlation features by scanning the three-dimensional feature map with a 3x3 convolutional kernel and outputs a convolutional feature map. The activation layer performs ReLU nonlinear activation on the convolutional feature map and outputs an activated feature map. The max-pooling layer performs 2x2 window maximum downsampling on the activated feature map and outputs a down-sampled feature map. The flattening layer flattens the down-sampled feature map into a renewable energy output feature vector. The fusion output submodule comprises: The feature concatenation layer concatenates the electricity price fluctuation trend vector, the load change trend vector, and the renewable energy output feature vector into a fusion feature vector. The full connection regression layer performs weighted calculation on the fusion feature vector and outputs the peak-valley electricity price difference mean value, the industrial and commercial user electricity load fluctuation amplitude, and the renewable energy output prediction value in a specified time period. The random disturbance branch layer adds Gaussian noise to the renewable energy output prediction value and then generates a renewable energy output confidence interval through Monte Carlo sampling.

[0064] The LSTM of the time sequence feature extraction submodule processes historical electricity price and load records and outputs the electricity price fluctuation trend vector and the load change trend vector, the CNN of the spatial feature extraction submodule processes renewable energy output records and generates the output feature vector, the fusion output submodule concatenates the feature vectors and then outputs the prediction indicators through the full connection regression layer, the random disturbance branch layer adds Gaussian noise and then generates the confidence interval through Monte Carlo sampling, and a prediction model architecture with multiple modules is constructed.

[0065] In some specific embodiments, the random disturbance branch layer adds Gaussian noise with a mean of 0 and a standard deviation of the fluctuation rate of the renewable energy output historical records to the renewable energy output prediction value, and then generates a renewable energy output confidence interval with a confidence level of 90% through 1000 times of Monte Carlo sampling.

[0066] In some specific embodiments, the training method of the electricity trading index prediction model comprises: The training set and the test set are constructed using the electricity trading data of the past three years, and are divided according to the time sequence in a 7:3 ratio. The training process adopts an end-to-end supervised learning manner: the standardized historical parameter set is taken as an input feature, and the actual recorded peak-valley price difference mean, industrial and commercial user electricity load fluctuation amplitude, and renewable energy output value in the corresponding period are taken as training labels; The loss function adopts a mean square error calculation manner, and the mathematical expression is the square difference mean of the predicted value and the actual value; The optimizer selects the Adam adaptive matrix estimation algorithm, the initial learning rate is set to 0.001, and the number of batch training samples is fixed to 64; The training termination condition is set to automatically stop iteration when the loss of the validation set decreases by less than 0.1% for five consecutive training periods, to avoid overfitting.

[0067] In some specific embodiments, formulating a power transaction plan based on power transaction prediction indicators includes the following steps: S301. Construct a multi-objective optimization function:

[0068] Wherein, is the peak-valley price difference mean in the time period output by the power transaction indicator prediction model; is the predicted total electricity load in the time period , which is equal to the average electricity load per unit time multiplied by the length of the time period, wherein the average electricity load per unit time is calculated based on the historical records of industrial and commercial user electricity load; is the capacity leasing revenue, , is the leased capacity, is the preset unit price of capacity leasing; is the ancillary service revenue, , is the ancillary service proportion coefficient, is the preset unit price of ancillary service; is the green electricity transaction volume; is the preset green electricity premium rate; , , is the weight coefficient, ; S302. Set three output scenarios based on the new energy processing confidence interval: Pessimistic scenario, pessimistic scenario output , set the optimistic probability weight to ; reference scenario, reference scenario output , set the reference probability weight as ; optimistic scenario, optimistic scenario output , set the optimistic probability weight as ; satisfy: ; S303. The multi-objective optimization function is taken as the objective function of the stochastic dual dynamic programming, and the decision variables including the power transaction volume , the leasing capacity , and the green power transaction volume are solved by the stochastic dual dynamic programming. The solution of the power transaction volume is used to optimize the component, and the calculation formula is as follows:

[0069] is the fluctuation amplitude of the industrial and commercial user electricity load output by the power transaction index prediction model; is a preset peak-valley adjustment coefficient; is the new energy penetration rate, which is the ratio of the historical average processing of renewable energy to the historical average electricity load of industrial and commercial users; The solution of the leasing capacity is used to optimize the component, and the calculation formula is as follows:

[0070] is a preset confidence correction factor; The solution of the green power transaction volume is used to optimize the component, and the calculation formula is as follows:

[0071] S304. The power transaction plan is output, and the power transaction plan includes: a power contract scheme including the power transaction volume , the power transaction price , , is a preset price adjustment coefficient; a capacity leasing scheme including the leasing capacity , a leasing period, and the leasing period is equal to the time period ; a green power transaction scheme including the green power transaction volume , green electricity premium rate .

[0072] By constructing a multi-objective optimization function with the mean of peak-valley electricity price difference, predicted total electricity load, capacity leasing income, ancillary service income and green electricity premium income as variables, combined with the three-scenario output values ​​and probability weights set based on confidence intervals, stochastic dual dynamic programming is used to solve the electricity trading volume, leasing capacity and green electricity trading volume, forming a mathematical optimization framework for multi-objective risk decision-making.

[0073] Step S4, converting the power trading plan into real-time constraints, inputting the power trading prediction indicators and the standardized power trading dynamic parameter set into the pre-trained trading strategy generation model, and generating charging and discharging power instructions that meet the real-time constraints.

[0074] By converting the power trading plan into real-time constraints, and inputting the power trading prediction indicators and the standardized power trading dynamic parameter set into the pre-training model to generate charging and discharging power instructions, it is ensured that the instruction output meets the constraint requirements of the trading plan conversion.

[0075] In some specific embodiments, the real-time constraints include: Charge and discharge power upper limit constraints:

[0076]

[0077] in, for Charging power or discharging power during the time period; for The upper limit of charge and discharge power in the time period; For the time period Hours expressed; is the preset charge and discharge efficiency coefficient; State of charge reservation constraints:

[0078]

[0079] in, for State of charge during the time period; The lowest reserved state of charge; is the rated capacity of the energy storage system; Daily green power trading volume limit constraints:

[0080]

[0081] For the Green electricity trading volume per day; The upper limit of daily green power trading volume; For the time period The number of days is expressed.

[0082] By defining the upper limit constraint of charge and discharge power to link the power trading volume with charge and discharge efficiency, the charge state reservation constraint to link the leased capacity with the rated capacity, and the upper limit constraint of daily green power trading volume to link the total green power volume with the number of days in the cycle, the trading plan is quantified into executable real-time control boundaries.

[0083] In some specific embodiments, the real-time constraints further include: Energy storage physical constraints: The upper limit of charge and discharge power is 90% of the rated power of the energy storage device, and the state of charge range is 20% to 90%; Grid operation constraints: voltage fluctuation range does not exceed ±5% of the nominal voltage, and frequency fluctuation range does not exceed ±0.2Hz of the nominal frequency; Trading rules and constraints: The electricity volume of a single transaction is 1 to 10MWh, and the trading time window is 1 hour before the opening of the day-ahead market to the closing of the market.

[0084] By supplementing the physical constraints of energy storage to limit the charging and discharging power limits and the range of charge status, the grid operation constraints to limit the voltage and frequency fluctuation thresholds, and the trading rule constraints to limit the trading power and time windows, the coverage dimension of real-time constraints is expanded.

[0085] In some specific embodiments, the trading strategy generation model includes a medium- and long-term planning submodule and a real-time decision optimization submodule, wherein: The medium- and long-term planning submodule is constructed using a long short-term memory network, including: Dynamic parameter input layer: receives power transaction prediction indicators and combines the power transaction prediction indicators of N consecutive specified time periods into a multi-period prediction indicator group; Strategy generation long short-term memory network layer: includes a strategy generation forget gate, a strategy generation input gate, and a strategy generation output gate. The strategy generation forget gate selects the long-term trend characteristics of the multi-period prediction indicator group, the strategy generation input gate extracts the short-term fluctuation characteristics of the multi-period prediction indicator group, and the strategy generation output gate generates the prediction indicator hidden state vector; Fully connected output layer: reduces the dimension of the hidden state vector of the prediction indicator and outputs the medium- and long-term charging and discharging plan; The real-time decision optimization submodule is constructed using a proximal policy optimization algorithm, including: a state input layer that receives a standardized power transaction dynamic parameter set, a medium- and long-term discharge plan, and real-time constraint conditions, and then performs feature splicing to form a state vector; a policy network layer that receives the state vector and calculates an action probability distribution through a fully connected neural network; a value network layer that receives the state vector and calculates a state value through a fully connected neural network; an action output layer that receives the action probability distribution and the state value, selects an optimal action through a proximal policy optimization algorithm, and outputs a charging and discharging power instruction that satisfies the real-time constraint conditions.

[0086] The multi-period prediction indicators are combined into a medium- and long-term charging and discharging plan through the medium- and long-term planning submodule, the dynamic parameter set, the medium- and long-term plan, and the real-time constraints are spliced into a state vector through the real-time decision optimization submodule, the action probability distribution is output through the policy network, the state value is evaluated through the value network, and finally a charging and discharging instruction that satisfies the constraints is generated, realizing the cooperative optimization of long- and short-term strategies.

[0087] In some specific embodiments, the medium- and long-term charging and discharging plan output by the medium- and long-term planning submodule is a structured vector :

[0088] wherein: is the number of hours in the time period . represents the target charging and discharging power baseline value in the future hour, when discharging, when charging.

[0089] By defining the medium- and long-term charging and discharging plan as a structured vector and clearly defining the meanings of positive and negative values of charging and discharging power, the data format of the standardized plan output is In some specific embodiments, in the process of generating the charging and discharging power instruction by the real-time decision optimization submodule using the proximal policy optimization algorithm, to ensure that the output action strictly satisfies the real-time constraint conditions, a double-path constraint satisfaction mechanism is implemented: The first path is an action mask mechanism. Before the candidate action selection in the action output layer, based on the real-time collected power grid operation data and the energy storage system state, the allowable charging and discharging power boundary value and the state of charge safety threshold in the current period are calculated, and a binary action mask vector is generated. This vector directly screens out illegal action options that exceed the power upper limit, violate the SOC lower limit, or break the power grid frequency fluctuation limit, ensuring that the action set evaluated by the policy network is completely compliant. The second path is a reward function penalty mechanism, which introduces a dynamic penalty term in the state value calculation of the value network. When the predicted action may cause the voltage deviation to exceed the nominal value ±5% or the frequency deviation to exceed ±0.2Hz, a quadratic penalty weight is added in the reward function, and the penalty coefficient is in a square relationship with the deviation amount, which encourages the policy network to actively avoid critical risk states.

[0090] This dual mechanism combines physical constraint hard truncation with target function soft guidance to achieve closed-loop satisfaction of real-time constraints.

[0091] In some specific embodiments, the transaction strategy generation model adopts a hierarchical progressive strategy for training, including: In the training of the medium and long term planning submodule, a multi-period prediction index group is used as the input feature, and the historical optimal charging and discharging power sequence is used as the supervision label. A smooth L1 loss function is used for parameter optimization. The real-time decision optimization submodule training adopts a proximal policy optimization algorithm, and its core reward function includes three components: energy revenue, grid stability, and constraint violation penalty. The energy revenue item quantifies the price difference revenue generated by charging and discharging behavior. The grid stability item rewards the frequency fluctuation control effect in an exponential decay form. The constraint violation penalty item imposes a linear penalty on the state of charge out-of-limit behavior. The training parameter configuration includes: the future revenue discount factor is set to 0.99, the strategy network parameter update step is set to 0.002, and the number of parallel training environments is set to 8 to accelerate convergence.

[0092] Step S5, the energy storage operator executes the real-time charging and discharging power instruction.

[0093] By executing the real-time charging and discharging power instruction by the energy storage operator, the physical execution loop of the power transaction strategy is completed.

[0094] In some specific embodiments, real-time monitoring and feedback are performed during the execution of the charging and discharging power instruction, including: Real-time acquisition of grid frequency data and energy storage device operating state; When the grid frequency fluctuation exceeds ±0.2Hz or the energy storage device temperature exceeds 60℃, an abnormal handling mechanism is triggered; The abnormal handling mechanism includes: suspending the current charging and discharging instruction, re-solving the transaction strategy generation model to generate an adjusted instruction, and issuing and executing it within 10 seconds.

[0095] By monitoring the grid frequency and energy storage temperature when executing the instruction, the instruction is suspended and an adjusted instruction is generated when the limit is exceeded, establishing a rapid response mechanism for abnormal working conditions.

[0096] In one specific embodiment, the steps of the power transaction optimization method for industrial and commercial energy storage include: Step S1, real-time collection of power transaction dynamic parameters and acquisition of power transaction historical characteristic data; The power transaction dynamic parameters include: The power grid price signal and power grid operation data provided by the power grid company, the power grid operation data including the load, voltage and frequency of the power grid; The power consumption load data and energy storage system state of charge data provided by the industrial and commercial users, the energy storage system state of charge data including the charge and discharge power, temperature and voltage of the energy storage system; The energy storage device operating state parameters provided by the energy storage operator; The renewable energy supply data provided by the renewable energy supplier; The power transaction historical characteristic data includes: Historical price records; Historical records of industrial and commercial user power consumption load; Renewable energy output historical records; The power transaction dynamic parameters and power transaction historical characteristic data are securely shared and credibly verified through blockchain or smart contract technology.

[0097] Step S2, data filtering, missing value filling and outlier correction are performed on the power transaction dynamic parameters to obtain a standardized power transaction dynamic parameter set; The missing value filling and normalization are performed on the transaction historical characteristic data to generate a standardized historical parameter set.

[0098] Step S3, the standardized historical parameter set is input into the pre-trained power transaction index prediction model to generate power transaction prediction indexes within a specified time period, and a power transaction plan for the same time period is formulated based on the power transaction prediction indexes; The power transaction prediction indexes include the average peak-valley price difference, the industrial and commercial user power consumption load fluctuation amplitude, and the renewable energy output confidence interval; The power transaction index prediction model includes a time series feature extraction submodule, a spatial feature extraction submodule, and a fusion output submodule, wherein: The time series feature extraction submodule is constructed based on a long short-term memory network and includes: A time series feature input layer: receiving time series data of historical price records and industrial and commercial user power consumption load historical records; A feature extraction long short-term memory network layer: including a feature extraction forget gate, a feature extraction input gate, and a feature extraction output gate, the feature extraction forget gate filters long-term trend features of historical prices, the feature extraction input gate extracts short-term correlation features of industrial and commercial user power consumption load fluctuations, and the feature extraction output gate generates a time series feature hidden state vector; The full connection compression layer includes a first branch and a second branch, the first branch reduces dimensionality of the time sequence feature hidden state vector to output the electricity price fluctuation trend vector, and the second branch reduces dimensionality of the time sequence feature hidden state vector to output the electricity load change trend vector; The spatial feature extraction submodule is constructed based on a convolutional neural network and includes: The spatial feature input layer receives the renewable energy output historical record, and organizes the renewable energy output historical record into a three-dimensional feature map with a width, a height and a number of channels according to the meteorological station where the renewable energy power generation equipment is located; The convolutional layer extracts regional meteorological correlation features by scanning the three-dimensional feature map through a 3*3 convolutional kernel, and outputs a convolutional feature map; The activation layer performs ReLU nonlinear activation on the convolutional feature map, and outputs an activated feature map; The max-pooling layer performs 2*2 window maximum downsampling on the activated feature map, and outputs a down-sampled feature map; The flattening layer flattens the down-sampled feature map into a renewable energy output feature vector; The fusion output submodule includes: The feature concatenation layer concatenates the electricity price fluctuation trend vector, the load change trend vector and the renewable energy output feature vector into a fusion feature vector; The full connection regression layer performs weighted calculation on the fusion feature vector, and outputs the average peak-valley electricity price difference, the industrial and commercial user electricity load fluctuation amplitude and the renewable energy output prediction value in a specified time period; The random disturbance branch layer adds Gaussian noise with a mean of 0 and a standard deviation of the renewable energy output historical record fluctuation rate to the renewable energy output prediction value, and then generates a renewable energy output confidence interval with a confidence level of 90% through 1000 times of Monte Carlo sampling; The power transaction plan based on the power transaction prediction index includes the following steps: S301. Construct a multi-objective optimization function:

[0099] Wherein, is the average peak-valley electricity price difference in the time period output by the power transaction index prediction model; is the predicted total electricity load in the time period , which is equal to the product of the unit time average electricity load and the time period length, wherein the unit time average electricity load is calculated based on the industrial and commercial user electricity load historical record; is the capacity leasing revenue, , is the leased capacity, a preset capacity leasing unit price; a secondary service income, , a secondary service proportionality coefficient, a preset secondary service unit price; a green electricity transaction volume; a preset green electricity premium rate; , , a weight coefficient, ; S302. Based on the new energy processing confidence interval, three output scenarios are set: a pessimistic scenario, the pessimistic scenario output , the optimistic probability weight is set to ; a benchmark scenario, the benchmark scenario output , the benchmark probability weight is set to ; an optimistic scenario, the optimistic scenario output , the optimistic probability weight is set to ; while satisfying: ; S303. The multi-objective optimization function is used as the objective function of the stochastic dual dynamic programming, and the decision variables are solved by the stochastic dual dynamic programming, including the power transaction volume , the leasing capacity , and the green electricity transaction volume ; wherein the solution of the power transaction volume is used to optimize the component, and the calculation formula is:

[0100] the fluctuation amplitude of the industrial and commercial user electricity load output by the power transaction index prediction model; a preset peak-valley adjustment coefficient; the new energy penetration rate, which is the ratio of the historical average processing of renewable energy to the historical average electricity load of industrial and commercial users; the solution of the leasing capacity is used to optimize the component, and the calculation formula is:

[0101] is the preset confidence correction factor; Green electricity trading volume The solution is used to optimize The calculation formula is:

[0102] S304. Output power trading plan. The power trading plan includes: Power contract schemes, including power trading volumes , electricity trading prices , , is the preset price adjustment factor; Capacity leasing solutions, including leasing capacity , Lease period, which is equal to the time period ; Green electricity trading scheme, including green electricity trading volume , green electricity premium rate .

[0103] Step S4: converting the power trading plan into real-time constraints, inputting the power trading prediction indicators and the standardized power trading dynamic parameter set into the pre-trained trading strategy generation model, and generating charging and discharging power instructions that meet the real-time constraints; Real-time constraints include: Charge and discharge power upper limit constraints:

[0104]

[0105] in, for Charging power or discharging power during the time period; for The upper limit of charge and discharge power in the time period; For the time period Hours expressed; is the preset charge and discharge efficiency coefficient; State of charge reservation constraints:

[0106]

[0107] in, for State of charge during the time period; The lowest reserved state of charge; is the rated capacity of the energy storage system; Daily green power trading volume limit constraints:

[0108]

[0109] For the Green electricity trading volume per day; The upper limit of daily green power trading volume; For the time period The number of days is expressed as; Energy storage physical constraints: The upper limit of charge and discharge power is 90% of the rated power of the energy storage device, and the state of charge range is 20% to 90%; Grid operation constraints: voltage fluctuation range does not exceed ±5% of the nominal voltage, and frequency fluctuation range does not exceed ±0.2Hz of the nominal frequency; Trading rules and constraints: The power volume of a single transaction is 1 to 10MWh, and the trading time window is from 1 hour before the opening of the day-ahead market to the closing of the market; The trading strategy generation model includes a medium- and long-term planning submodule and a real-time decision optimization submodule, where: The medium- and long-term planning submodule is constructed using a long short-term memory network, including: Dynamic parameter input layer: receives power transaction prediction indicators and combines the power transaction prediction indicators of N consecutive specified time periods into a multi-period prediction indicator group; Strategy generation long short-term memory network layer: includes a strategy generation forget gate, a strategy generation input gate, and a strategy generation output gate. The strategy generation forget gate selects the long-term trend characteristics of the multi-period prediction indicator group, the strategy generation input gate extracts the short-term fluctuation characteristics of the multi-period prediction indicator group, and the strategy generation output gate generates the prediction indicator hidden state vector; Fully connected output layer: Reduce the dimension of the hidden state vector of the prediction indicator and output the medium- and long-term charging and discharging plan. The medium- and long-term charging and discharging plan is a structured vector :

[0110] in: Time period Time period the number of hours; Indicates the future The target charge and discharge power baseline value for the hour, When it indicates discharge, When it is charging; The real-time decision optimization submodule is constructed using a proximal policy optimization algorithm, and includes: a state input layer: receiving a set of standardized power transaction dynamic parameters, a medium and long-term discharge plan, and real-time constraint conditions, and then performing feature splicing to form a state vector; a policy network layer: receiving the state vector, and calculating an action probability distribution through a fully connected neural network; a value network layer: receiving the state vector, and calculating a state value through a fully connected neural network; an action output layer: receiving the action probability distribution and the state value, selecting an optimal action through a proximal policy optimization algorithm, and outputting a charging and discharging power instruction that satisfies the real-time constraint condition.

[0111] Step S5: The energy storage operator executes the real-time charging and discharging power instruction, and monitors and feeds back in real time during the execution of the charging and discharging power instruction, including: real-time collection of power grid frequency data and energy storage device operating state; when it is detected that the power grid frequency fluctuation exceeds ±0.2 Hz or the energy storage device temperature exceeds 60℃, an abnormal processing mechanism is triggered; the abnormal processing mechanism includes: suspending the current charging and discharging instruction, re-solving the transaction strategy generation model to generate an adjusted instruction, and issuing and executing the adjusted instruction within 10 seconds.

[0112] The following is an embodiment of the power transaction optimization system for industrial and commercial energy storage provided by the embodiment, which belongs to the same inventive concept as the power transaction optimization method of each of the above embodiments. Details not described in detail in the embodiment of the power transaction optimization system can be referred to the above embodiment of the power transaction optimization method for industrial and commercial energy storage.

[0113] As shown in Figure 2 the power transaction optimization system for industrial and commercial energy storage includes: a data acquisition module for real-time acquisition of power transaction dynamic parameters and acquisition of power transaction historical feature data; a data preprocessing module for preprocessing the power transaction dynamic parameters and transaction historical feature data to obtain a set of standardized power transaction dynamic parameters and a set of standardized historical parameters; an index prediction and plan making module for inputting the set of standardized historical parameters into a pre-trained power transaction index prediction model to generate power transaction prediction indexes within a specified time period, and making a power transaction plan for the same time period based on the power transaction prediction indexes; an instruction generation module for converting the power transaction plan into real-time constraint conditions, inputting the power transaction prediction indexes and the set of standardized power transaction dynamic parameters into a pre-trained transaction strategy generation model, and generating a charging and discharging power instruction that satisfies the real-time constraint condition. The execution module is configured to execute the real-time charging and discharging power instruction.

[0114] The power transaction optimization system of the embodiment is used to implement the power transaction optimization method for industrial and commercial energy storage.

[0115] The application further provides an electronic device for implementing various embodiments of the application, Figure 3 A hardware structure schematic diagram of an electronic device for implementing various embodiments of the application is shown in the figure. Figure 3 As shown in the figure, the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0116] Those skilled in the art can understand that the electronic device structure involved in the embodiments of the application does not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than the illustration, or combine certain components, or different component arrangements.

[0117] In the embodiments of the application, the electronic device includes but is not limited to a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the application described and / or claimed herein.

[0118] In addition, the electronic device includes some functional modules that are not shown and will not be described here.

[0119] The application further provides a storage medium in which a program product capable of implementing the power transaction optimization method for industrial and commercial energy storage is stored. In some possible implementation manners, various aspects of the application can also be implemented in the form of a program product, which includes program code for causing the terminal device to execute the steps according to various exemplary embodiments of the application described in the "Exemplary Method" section of the specification when the program product is run on the terminal device.

[0120] The storage medium can be any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0121] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing power trading for industrial and commercial energy storage, characterized in that: include: S1. Real-time collection of dynamic parameters of power transactions and acquisition of historical characteristic data of power transactions; S2. Preprocessing the power trading dynamic parameters and historical transaction characteristic data to obtain a standardized power trading dynamic parameter set and a standardized historical parameter set; S3. Input the standardized historical parameter set into the pre-trained power trading indicator prediction model to generate power trading prediction indicators for a specified time period, and formulate a power trading plan for the same time period based on the power trading prediction indicators; Power trading forecast indicators include the average difference between peak and valley electricity prices, the fluctuation range of industrial and commercial user power load, and the confidence interval of renewable energy output; S4. Convert the power trading plan into real-time constraints, input the power trading prediction indicators and the standardized power trading dynamic parameter set into the pre-trained trading strategy generation model, and generate charging and discharging power instructions that meet the real-time constraints; S5. The energy storage operator executes real-time charging and discharging power instructions.

2. The power transaction optimization method according to claim 1, characterized in that: In step S1, the dynamic parameters of power trading include: Grid price signals and grid operation data provided by the grid company, including grid load, voltage, and frequency; Power load data and energy storage system state of charge data provided by industrial and commercial users. The energy storage system state of charge data includes the energy storage system's charge and discharge power, temperature, and voltage; Energy storage equipment operating status parameters provided by the energy storage operator; Renewable energy output data provided by renewable energy suppliers; The historical characteristic data of power transactions include: Historical electricity price records; Historical records of electricity load for industrial and commercial users; Historical records of renewable energy output.

3. The power transaction optimization method according to claim 1, characterized in that: Step S2 includes: Perform data filtering, missing value filling and outlier correction on the power transaction dynamic parameters to obtain a standardized power transaction dynamic parameter set; Fill missing values ​​and normalize transaction history feature data to generate a standardized historical parameter set.

4. The power transaction optimization method according to claim 2, characterized in that: In step S3, the power trading index prediction model includes a time series feature extraction submodule, a spatial feature extraction submodule, and a fusion output submodule, wherein: The temporal feature extraction submodule is built based on the long short-term memory network and includes: Time series feature input layer: receives time series data of historical electricity price records and electricity load history records of industrial and commercial users; Feature extraction long-term short-term memory network layer: includes feature extraction forget gate, feature extraction input gate and feature extraction output gate. The feature extraction forget gate selects the long-term trend characteristics of historical electricity prices. The feature extraction input gate extracts the short-term correlation characteristics of industrial and commercial users' electricity load fluctuations. The feature extraction output gate generates the time series feature hidden state vector. Fully connected compression layer: It includes the first branch and the second branch. The first branch reduces the dimension of the hidden state vector of the time series feature and outputs the electricity price fluctuation trend vector. The second branch reduces the dimension of the hidden state vector of the time series feature and outputs the electricity load change trend vector. The spatial feature extraction submodule is built based on a convolutional neural network and includes: Spatial feature input layer: Receives the historical record of renewable energy output and organizes it into a three-dimensional feature map of width × height × number of channels according to the distribution of meteorological stations where renewable energy power generation equipment is located; Convolution layer: Scans the 3D feature map with a 3×3 convolution kernel to extract regional meteorological related features and outputs a convolution feature map; Activation layer: Perform ReLU nonlinear activation on the convolution feature map and output the activation feature map; Maximum pooling layer: performs maximum downsampling of the activation feature map in a 2×2 window and outputs the downsampled feature map; Flattening layer: flattens the downsampled feature map into a renewable energy output feature vector; The fusion output submodule includes: Feature splicing layer: Splices the electricity price fluctuation trend vector, load change trend vector, and renewable energy output feature vector into a fused feature vector; Fully connected regression layer: This layer performs weighted calculations on the fused feature vectors and outputs the average peak-valley price difference, the fluctuation amplitude of industrial and commercial user electricity load, and the predicted output value of renewable energy within a specified time period. Random perturbation branch layer: Gaussian noise is added to the renewable energy output prediction value, and then the renewable energy output confidence interval is generated through Monte Carlo sampling.

5. The power transaction optimization method according to claim 2, characterized in that: In step S3, formulating a power trading plan based on the power trading forecast indicator includes the following steps: S301. Constructing a multi-objective optimization function: in, The time period for the output of the power trading index prediction model The average difference between peak and valley electricity prices within ; For the time period The total electricity load forecasted within the period is equal to the product of the average electricity load per unit time and the length of the time period, where the average electricity load per unit time is calculated based on the historical records of electricity load for industrial and commercial users; For capacity leasing income, , For leasing capacity, The unit price for the preset capacity rental; For ancillary service income, , is the auxiliary service ratio coefficient, is the preset unit price of auxiliary services; Green electricity trading volume; is the preset green electricity premium rate; 、 、 is the weight coefficient, ; S302. Set three output scenarios based on the confidence interval of new energy processing: Pessimistic scenario, pessimistic scenario output , set the optimistic probability weight to ; Baseline scenario, baseline scenario output , set the baseline probability weight to ; Optimistic scenario, optimistic scenario output , set the optimistic probability weight to ; At the same time: ; S303. The multi-objective optimization function is used as the objective function of the stochastic dual dynamic programming, and the decision variables are solved by the stochastic dual dynamic programming. The decision variables include the power trading volume. , rental capacity , green electricity trading volume ; Among them, the electricity trading volume The solution is used to optimize The calculation formula is: The fluctuation range of industrial and commercial user electricity load output by the power trading index prediction model; is the preset peak-valley adjustment coefficient; is the new energy penetration rate, which is the ratio of the historical average treatment of renewable energy to the historical average electricity load of industrial and commercial users; Leased capacity The solution is used to optimize The calculation formula is: is the preset confidence correction factor; Green electricity trading volume The solution is used to optimize The calculation formula is: S304. Output power trading plan. The power trading plan includes: Power contract schemes, including power trading volumes , electricity trading prices , , is the preset price adjustment factor; Capacity leasing solutions, including leasing capacity , Lease period, which is equal to the time period ; Green electricity trading scheme, including green electricity trading volume , green electricity premium rate .

6. The power transaction optimization method according to claim 5, characterized in that: In step S4, the real-time constraints include: Charge and discharge power upper limit constraints: in, for Charging power or discharging power during the time period; for The upper limit of charge and discharge power in the time period; For the time period Hours expressed; is the preset charge and discharge efficiency coefficient; State of charge reservation constraints: in, for State of charge during the time period; The lowest reserved state of charge; is the rated capacity of the energy storage system; Daily green electricity trading volume limit constraints: For the Green electricity trading volume per day; The upper limit of daily green power trading volume; For the time period The number of days is expressed.

7. The power transaction optimization method according to claim 6, characterized in that: In step S4, the real-time constraints also include: Energy storage physical constraints: The upper limit of charge and discharge power is 90% of the rated power of the energy storage device, and the state of charge range is 20% to 90%; Grid operation constraints: voltage fluctuation range does not exceed ±5% of the nominal voltage, and frequency fluctuation range does not exceed ±0.2Hz of the nominal frequency; Trading rules and constraints: The electricity volume of a single transaction is 1 to 10MWh, and the trading time window is 1 hour before the opening of the day-ahead market to the closing of the market.

8. The power transaction optimization method according to claim 1, characterized in that: In step S4, the trading strategy generation model includes a medium- and long-term planning submodule and a real-time decision optimization submodule, wherein: The medium- and long-term planning submodule is constructed using a long short-term memory network, including: Dynamic parameter input layer: receives power transaction prediction indicators and combines the power transaction prediction indicators of N consecutive specified time periods into a multi-period prediction indicator group; Strategy generation long short-term memory network layer: includes a strategy generation forget gate, a strategy generation input gate, and a strategy generation output gate. The strategy generation forget gate selects the long-term trend characteristics of the multi-period prediction indicator group, the strategy generation input gate extracts the short-term fluctuation characteristics of the multi-period prediction indicator group, and the strategy generation output gate generates the prediction indicator hidden state vector; Fully connected output layer: reduces the dimension of the hidden state vector of the prediction indicator and outputs the medium- and long-term charging and discharging plan; The real-time decision optimization submodule is built using a proximal strategy optimization algorithm, including: State input layer: Receives standardized power trading dynamic parameter sets, medium- and long-term discharge plans, and real-time constraints, then performs feature concatenation to form a state vector; Policy network layer: receives the state vector and calculates the action probability distribution through a fully connected neural network; Value network layer: receives the state vector and calculates the state value through a fully connected neural network; Action output layer: Receives action probability distribution and state value, selects the optimal action through the proximal strategy optimization algorithm, and outputs charging and discharging power instructions that meet real-time constraints.

9. The power transaction optimization method according to claim 1, characterized in that: Step S5 performs real-time monitoring and feedback during the execution of the charge and discharge power instructions, including: Real-time collection of grid frequency data and energy storage equipment operating status; When the grid frequency fluctuation exceeds ±0.2Hz or the energy storage device temperature exceeds 60°C, the abnormal handling mechanism is triggered; The exception handling mechanism includes: suspending the current charge and discharge instructions, re-solving the trading strategy generation model to generate adjusted instructions, and issuing them for execution within 10 seconds.

10. An electricity trading optimization system for industrial and commercial energy storage, characterized in that: A method for implementing the power transaction optimization method according to any one of claims 1 to 9, comprising: Data acquisition module, used to collect dynamic parameters of power transactions in real time and obtain historical characteristic data of power transactions; A data preprocessing module is used to preprocess the power transaction dynamic parameters and transaction history characteristic data to obtain a standardized power transaction dynamic parameter set and a standardized historical parameter set; The indicator forecasting and planning module is used to input the standardized historical parameter set into the pre-trained power trading indicator forecasting model, generate power trading forecast indicators within a specified time period, and formulate a power trading plan for the same time period based on the power trading forecast indicators; The instruction generation module is used to convert the power trading plan into real-time constraints, input the power trading prediction indicators and standardized power trading dynamic parameter set into the pre-trained trading strategy generation model, and generate charging and discharging power instructions that meet the real-time constraints; The execution module is used to execute real-time charge and discharge power instructions.

Images