A transformer station multi-energy complementary thermal management system and a collaborative regulation method thereof

Abstract

The application discloses a transformer substation multi-energy complementary heat management system and a collaborative regulation method thereof, and belongs to the field of automatic control technology. The system comprises the following steps: acquiring meteorological environment prediction data, historical operation data and real-time energy state data, performing time-frequency domain feature extraction and fusion, generating a prediction input feature vector, inputting the heat load prediction model, and obtaining a heat load demand prediction sequence; acquiring dynamic response characteristic parameters, constructing a device dynamic response matrix, performing multi-objective rolling time domain optimization, and obtaining a pre-scheduling plan sequence; acquiring heat delay characteristic parameters and the capacity state of a heat storage device, correcting the pre-scheduling plan sequence into an execution scheduling instruction, and generating control signals for each energy conversion device and the heat storage device. The collaborative regulation strategy based on multi-dimensional feature prediction, dynamic constraint optimization and closed-loop feedback correction can realize the collaborative optimization scheduling of multi-energy devices and improve the economy and stability of system operation.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to a multi-energy complementary thermal management system for substations and its coordinated control method. Background Technology

[0002] As the core hub of the power system, substations generate enormous amounts of heat during the operation of their numerous electrical devices. To ensure the safe and stable operation of these devices, effective thermal management is essential. The multi-energy complementary thermal management system for substations integrates various energy sources such as natural gas, electricity, solar energy, and geothermal energy, utilizing boilers, heat pumps, and thermal storage devices for comprehensive heating or cooling. This is a crucial technological approach for achieving cascaded energy utilization and improving energy efficiency.

[0003] Among related technologies, Chinese invention patent CN111967786B discloses a hierarchical collaborative control method for multi-energy complementary microgrids, including: constructing an integrated comprehensive model that takes into account both the overall multi-energy complementary microgrid and the different interests of the various energy subsystems (electricity, heat, cooling, and gas) within the multi-energy complementary microgrid, guiding the collaborative planning and optimization control research of the multi-energy complementary microgrid; dividing the decision variables of the four types of energy (electricity, heat, cooling, and gas) into two layers based on their inertia, transmission speed, and response time of the energy conversion equipment, achieving refined energy management; and adjusting the decision variables in the two layers at different time intervals according to the characteristics of the four types of energy and the integrated comprehensive model, gradually reducing the impact of source-load uncertainty on the system and improving the quality of energy control in the system.

[0004] However, the hierarchical logic of the aforementioned technologies relies on static classification, fails to integrate real-time operational and predictive information, lacks a clear quantitative method for assessing heat load change trends, and lacks online identification of equipment dynamic responses. This results in fixed boundary conditions, making it difficult to adapt to frequent fluctuations caused by weather and equipment conditions. Furthermore, its utilization of thermal inertia is limited to time-scale divisions, without establishing mathematical constraints on heat transfer delays and thermal storage capacity. This weakens the uncertainty mitigation effect and easily leads to supply-demand mismatches and energy curtailment problems. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a multi-energy complementary thermal management system for substations and its collaborative control method. It employs a collaborative control strategy based on multi-dimensional feature prediction, dynamic constraint optimization, and closed-loop feedback correction, which enables collaborative optimization scheduling of multiple energy devices and improves the economy and stability of system operation.

[0006] The above objectives can be achieved through the following approach:

[0007] A method for coordinated control of thermal management in a substation with multi-energy complementarity includes: acquiring meteorological and environmental forecast data of the substation location, historical operating data of the substation's heat load, and real-time energy status data of the multi-energy complementarity system; extracting time-frequency domain features from the meteorological and environmental forecast data and the historical operating data, and fusing them with the real-time energy status data to generate a predictive input feature vector; inputting the predictive input feature vector into a pre-constructed heat load prediction model to obtain a heat load demand prediction sequence for the future control period; acquiring dynamic response characteristic parameters of each energy conversion device in the multi-energy complementarity system, and constructing a device dynamic response matrix based on the dynamic response characteristic parameters; using the heat load demand prediction sequence as the tracking target, and combining it with the device dynamic response matrix, performing multi-objective rolling time-domain optimization to obtain a pre-scheduling plan sequence; acquiring thermal delay characteristic parameters of the heat transmission network and the capacity status of the heat storage device, modifying the pre-scheduling plan sequence into an execution scheduling instruction that satisfies thermal inertia constraints, and generating control signals for each energy conversion device and heat storage device based on the execution scheduling instruction, and issuing them for execution.

[0008] Optionally, generating the prediction input feature vector includes: acquiring the time series of temperature, irradiance, and wind speed from the meteorological environment forecast data, and the heat load power series corresponding to the time series of temperature, irradiance, and wind speed from the historical operating data; performing wavelet transform on the time series of temperature, irradiance, and wind speed and the heat load power series to extract multi-scale time-frequency feature components; and concatenating and reducing the dimensionality of the multi-scale time-frequency feature components and the real-time energy state data to generate the prediction input feature vector.

[0009] Optionally, obtaining the heat load demand prediction sequence for the future control period includes: acquiring the heat load dataset of the substation for historical periods and the historical meteorological environment data corresponding to the historical periods; constructing a deep neural network model structure using an attention mechanism and a long short-term memory network; inputting the historical heat load dataset and the historical meteorological environment data into the deep neural network model structure for training to obtain a heat load prediction model; and combining the prediction input feature vector to output the heat load demand prediction sequence.

[0010] Optionally, constructing the device dynamic response matrix based on the dynamic response characteristic parameters includes: sending a step test signal to each energy conversion device and acquiring the response process curve of the device output power changing with time; performing parameter identification of the response process curve using a first-order inertial plus pure delay element to obtain the dynamic response time constant and ramp rate limit value of each energy conversion device; and combining the dynamic response time constant and the ramp rate limit value into a matrix to obtain the device dynamic response matrix.

[0011] Optionally, the process of obtaining the pre-scheduled plan sequence includes: constructing a system operating cost function that includes a renewable energy curtailment penalty term and an energy consumption term for the operation of each energy conversion device, as well as a renewable energy utilization rate target; using the heat load demand forecast sequence as a heat power balance constraint, using the device dynamic response matrix as a device output change rate constraint, and adding upper and lower power limit constraints for each energy conversion device; combining the system operating cost function and the renewable energy utilization rate target into a single objective function, performing rolling time-domain optimization to generate the pre-scheduled plan sequence.

[0012] Optionally, the construction of a system operating cost function including a renewable energy curtailment penalty term and an operating energy consumption term for each energy conversion device, as well as a renewable energy utilization rate target, includes: collecting the actual grid-connected power and the upper limit of available power during the same period of renewable energy power generation equipment, accumulating the hourly difference and superimposing a penalty weight that increases stepwise with the duration of curtailment to generate a renewable energy curtailment penalty term; obtaining the dynamic energy consumption characteristics of each energy conversion device during variable load operation, superimposing the transient energy consumption corresponding to the current output of the device with the additional energy consumption generated by start-stop switching to generate an operating energy consumption term; weighting the renewable energy curtailment penalty term and the operating energy consumption term according to the control cycle to generate a system operating cost function; obtaining the actual renewable energy power generation and available power generation for each time period, calculating the ratio for each time period, and then weighting and averaging the obtained total heat demand of the substation for each time period to generate a renewable energy utilization rate target.

[0013] Optionally, the step of modifying the pre-scheduled plan sequence into an execution scheduling instruction that satisfies thermal inertia constraints includes: extracting features based on the thermal delay characteristic parameters to obtain segment length parameters, pipe diameter parameters, and thermal resistance parameters of the insulation material, and calculating the pipe thermal delay time coefficient; extracting features based on the capacity status to obtain the current heat storage capacity and rated heat storage / discharge power value of the thermal storage device; establishing heat transfer delay constraints and thermal storage power capacity constraints based on the pipe thermal delay time coefficient and the rated heat storage / discharge power value; comparing and correcting the output timing of each device and the thermal storage charging / discharging plan in the pre-scheduled plan sequence with the heat transfer delay constraints and thermal storage power capacity constraints, and generating an execution scheduling instruction.

[0014] Optionally, the method further includes: acquiring the actual temperature signal at a designated hot node within the substation, calculating the temperature deviation between the actual temperature signal and the target temperature corresponding to the predicted heat load demand sequence; performing feedback compensation calculation based on the temperature deviation to generate a dynamic compensation power correction value; and feeding forward the dynamic compensation power correction value into the next round of execution scheduling instructions.

[0015] Optionally, the method further includes: obtaining real-time electricity price signals and ancillary service demand signals from the electricity market interface; incorporating the obtained real-time electricity price signals into the system operating cost function to calculate electricity purchase costs or electricity sales revenue, and simultaneously adding an ancillary service revenue term based on the ancillary service demand signals to update the system operating cost function; generating thermal storage charging and discharging arbitrage constraints adapted to electricity price fluctuations based on the capacity status and the real-time electricity price signals; and performing rolling time-domain optimization based on the updated system operating cost function and the thermal storage charging and discharging arbitrage constraints to obtain a combined thermal power pre-scheduling plan sequence coordinated with the electricity market signals.

[0016] Based on the same inventive concept, this invention also provides a multi-energy complementary thermal management system for substations. The system includes: a feature fusion module, used to acquire meteorological and environmental forecast data of the substation location, historical operating data of the substation's heat load, and real-time energy status data of the multi-energy complementary system; extract time-frequency domain features from the meteorological and environmental forecast data and the historical operating data; and fuse these features with the real-time energy status data to generate a predictive input feature vector; a load prediction module, used to input the predictive input feature vector into a pre-constructed heat load prediction model to obtain a predicted heat load demand sequence for future control periods; and a response matrix module. The system is used to acquire the dynamic response characteristic parameters of each energy conversion device in a multi-energy complementary system, and to construct a device dynamic response matrix based on the dynamic response characteristic parameters; the rolling optimization module is used to perform multi-objective rolling time-domain optimization with the heat load demand prediction sequence as the tracking target and the device dynamic response matrix to obtain the pre-scheduling plan sequence; the inertia correction module is used to acquire the thermal delay characteristic parameters of the heat transmission network and the capacity status of the heat storage device, correct the pre-scheduling plan sequence to an execution scheduling instruction that satisfies thermal inertia constraints, generate control signals for each energy conversion device and heat storage device based on the execution scheduling instruction, and issue them for execution.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] This invention integrates multi-dimensional time-frequency domain features of meteorological environmental forecast data, historical heat load operation data, and real-time energy status data, and constructs a high-precision heat load prediction model based on attention mechanism and long short-term memory network. This improves the ability to predict future heat load demand trends, enabling the control system to respond in advance to external environmental disturbances and changes in operating modes, and reducing the problem of supply and demand time scale mismatch.

[0019] This invention fully identifies the dynamic response time constant and ramp rate limit of each energy conversion device, constructs a device dynamic response matrix, and uses this matrix as a hard constraint on the output change rate in rolling time-domain optimization. At the same time, it introduces the delay characteristics of the heat transfer network and the state of heat storage capacity to correct the inertial constraint of the pre-scheduling plan, transforming the unfavorable thermal hysteresis effect into a virtual energy storage buffer, thereby ensuring that the scheduling command matches the actual physical capacity of the device and improving the device safety.

[0020] This invention addresses multiple objectives, including minimizing system operating costs and maximizing renewable energy utilization. It constructs a comprehensive optimization function that includes a deprecation penalty term and equipment energy consumption term. During the rolling optimization process, it combines thermal storage charging and discharging arbitrage strategy with real-time electricity market signals to achieve joint economic dispatch of thermal power. By calculating dynamic compensation power correction through closed-loop temperature feedback and feeding it forward to subsequent commands, it improves the local consumption rate of renewable energy, reduces operating costs, and enhances the adaptive robustness against various uncertainties.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims, and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a multi-energy complementary thermal management and coordinated control method for substations according to an embodiment of the present invention.

[0024] Figure 2 This is a graph showing the cumulative variance contribution rate of the feature matrix in a dimension reduction embodiment of the present invention.

[0025] Figure 3 This is a heat load demand prediction sequence curve diagram according to an embodiment of the present invention.

[0026] Figure 4 This is a bar chart showing the availability and consumption of renewable energy in different time periods according to an embodiment of the present invention.

[0027] Figure 5 This is a thermogram of the pipeline length and flow velocity versus the pipeline thermal delay coefficient according to an embodiment of the present invention.

[0028] Figure 6This is a schematic diagram of the structure of a multi-energy complementary thermal management system for a substation according to an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Reference Figure 1 One embodiment of the present invention proposes a multi-energy complementary thermal management collaborative control method for substations. It adopts a collaborative control strategy based on multi-dimensional feature prediction, dynamic constraint optimization and closed-loop feedback correction, which can realize the collaborative optimization scheduling of multiple energy devices and improve the economy and stability of system operation.

[0031] The method described in this embodiment specifically includes:

[0032] Acquire meteorological and environmental forecast data of the substation location, historical operating data of heat load within the substation, and real-time energy status data of the multi-energy complementary system. Extract time-frequency domain features from the meteorological and environmental forecast data and the historical operating data, and fuse them with the real-time energy status data to generate a predictive input feature vector.

[0033] Optionally, generating the predicted input feature vector includes:

[0034] Obtain the time series of temperature, irradiance, and wind speed from the meteorological environment forecast data, as well as the heat load power series corresponding to the time series of temperature, irradiance, and wind speed from the historical operating data;

[0035] Wavelet transform is performed on the time series of temperature, irradiance, and wind speed, and the heat load power series to extract multi-scale time-frequency feature components.

[0036] The multi-scale time-frequency feature components and the real-time energy state data are concatenated and their dimensions reduced to generate a predictive input feature vector.

[0037] Specifically, the process begins with acquiring three key types of data. The first type is meteorological and environmental forecast data, which is environmental prediction information about the substation over a future period obtained from meteorological service agencies. Specifically, this includes time series of temperature, solar irradiance, and wind speed at multiple future time points. The second type is historical operating data of the substation's heat load, extracted from the substation monitoring system or historical database. This includes records of heat load power consumption corresponding to past meteorological and environmental forecast data in time, forming a heat load power sequence. The third type is real-time energy status data of the multi-energy complementary system. This is a snapshot of the system's current operation collected in real time by sensors, such as the heat storage capacity of thermal storage equipment and the instantaneous output of renewable energy power generation equipment. After acquiring the above time series data, in order to reveal the deeper periodic patterns and abrupt changes in the data, feature extraction is required from the time series of temperature, irradiance, and wind speed, as well as the heat load power sequence. Wavelet transform technology is used here. It is a signal analysis tool that can analyze signals simultaneously in both time and frequency dimensions, and is particularly suitable for processing non-stationary signals containing abrupt changes. Through wavelet transform, each original time series is decomposed into a series of multi-scale time-frequency feature components. These components describe the details and overall trend of the original signal's variation at different time scales, i.e., frequencies. For calculating the wavelet coefficients at scale a and time b, this is a multi-scale time-frequency characteristic component. ,have:

[0038] ;

[0039] in, For any input time series, such as a temperature time series It is composed of the mother wavelet function. The wavelet basis functions generated after scaling and translation It is a conjugate form; The scaling factor is a positive real number that controls the scaling of the wavelet basis function. It is inversely proportional to the frequency and is used to analyze different frequency components of a signal. The range of values ​​can be set according to the analysis requirements. For example, to capture daily variation cycles and hourly fluctuations, a set of discrete values ​​can be set. The setting is based on the main frequency components determined after Fourier analysis of 200 sets of industrial sensor measured data. , where is a translation factor and is a real number, controls the translation of the wavelet basis function on the time axis, used to locate the time points where specific frequency components appear. By calculating for each time series at all set 'a' and 'b' values, a complete set of feature component matrices is obtained. After extracting the multi-scale time-frequency feature components of all time series, the next step is to integrate these high-dimensional feature information with real-time energy state data. The first step of integration is concatenation, which involves flattening all extracted multi-scale time-frequency feature component matrices into vectors and concatenating them at the end with the real-time energy state data, which is also in vector form, to form an initial feature vector. Since the concatenated vector has a high dimension, it may contain redundant information and increase the computational burden of subsequent models; therefore, dimensionality reduction is required. Dimensionality reduction aims to retain the most important information in the data while reducing the number of features. Through methods such as principal component analysis, the high-dimensional concatenated vector is projected into a low-dimensional space, ultimately generating the predicted input feature vector. The target dimension for dimensionality reduction is set to ensure that the reduced features can explain a high proportion of the variability in the original data. For example, based on the analysis of 500 sets of historical spliced ​​vectors, a target dimension value is selected to retain 95% of the original information. The final generated predictive input feature vector contains both deep time-frequency insights into the future environment and historical loads, and incorporates the current instantaneous state, providing a data foundation for prediction. Figure 2 As shown, with the increase of the number of principal components, the cumulative variance contribution rate exhibits a significant logarithmic saturation convergence trend. The figure uses a horizontal dashed line to set a rigid threshold of 95% information retention; the cumulative contribution rate curve first crosses this threshold when the number of principal components reaches 5.

[0040] For example, a forecast needs to be made for the next 4 hours (10:00, 11:00, 12:00, 13:00). Meteorological forecast data shows that the temperature time series for the next 4 hours is [18, 20, 21, 19] (unit: degrees Celsius), the irradiance time series is [600, 800, 850, 750] (unit: watts / square meter), and the wind speed time series is [2.1, 2.3, 2.0, 2.5] (unit: meters / second). Historical operational data shows that under similar meteorological conditions in the past, the corresponding heat load power series is [110, 130, 140, 125] (unit: kilowatts). The collected real-time energy status data is [0.7, 45], where 0.7 represents the current percentage of the thermal storage capacity of the thermal storage equipment (dimensionless), and 45 represents the current real-time output of the photovoltaic system (unit: kilowatts). Taking the temperature time series [18,20,21,19] as an example, applying discrete wavelet transform, such as using the Haar wavelet, after the first level of decomposition, yields approximate components (low-frequency trends) and detail components (high-frequency changes). The calculation process is as follows: Approximate component 1: Approximate component 2: Detail 1: Detail 2: After one layer of decomposition, the multi-scale time-frequency feature components of the temperature series can be represented as [26.87, 28.28, -1.41, 1.41]. Similarly, the same transformation is applied to the other three time series, yielding the feature components of the irradiance series as [989.9, 1131.4, -141.4, 70.7], the wind speed series as [3.11, 3.18, -0.14, -0.35], and the heat load power series as [169.7, 187.4, -14.1, 10.6]. The extracted feature components are then concatenated with the real-time energy state data. The four sets of feature components are concatenated with the real-time data vector to form a high-dimensional vector: [26.87,28.28,-1.41,1.41,989.9,1131.4,-141.4,70.7,3.11,3.18,-0.14,-0.35,169.7,187.4,-14.1,10.6,0.7,45]. The total dimension of this concatenated vector is [missing value]. Dimensionality reduction is targeted at 5 dimensions. This is based on the fact that, when performing principal component analysis on historical datasets, the first 5 principal components can explain more than 95% of the variance. The projection matrix obtained through principal component analysis is an 18x5 matrix, with the first row being [0.1, -0.02, 0.3, -0.15, 0.08, ...]. Through complete matrix multiplication, a 5-dimensional predicted input feature vector is finally obtained, such as [150.3, -25.1, 10.8, -5.2, 1.9].

[0041] The predicted input feature vector is input into a pre-built heat load prediction model to obtain a heat load demand prediction sequence for future control periods.

[0042] Optionally, obtaining the heat load demand forecast sequence for the future control period includes:

[0043] Obtain the heat load dataset of the substation for historical periods and the historical meteorological environment data corresponding to the historical periods;

[0044] Construct a deep neural network model structure using attention mechanisms and long short-term memory networks;

[0045] The historical heat load dataset and the historical meteorological environment data are input into the deep neural network model structure for training to obtain a heat load prediction model. Combined with the prediction input feature vector, a heat load demand prediction sequence is output.

[0046] Specifically, the first step is to build and train a heat load prediction model. Model building begins with data preparation, specifically acquiring historical heat load datasets for substations and corresponding historical meteorological data. The historical heat load dataset consists of a sequence of heat load power values ​​recorded at fixed time intervals, such as every hour, during the substation's past operation. The historical meteorological data consists of records of meteorological parameters such as temperature, irradiance, and wind speed that are completely synchronized with these heat load records in time. The data collection period needs to be long enough; for example, based on the analysis of substation operating characteristics, hourly data from two consecutive years can be used to cover complete seasonal, periodic, and holiday variation patterns. A deep neural network model structure is then constructed using attention mechanisms and Long Short-Term Memory (LSTM) networks. LTM networks are a special type of recurrent neural network that selectively remembers and forgets information through sophisticated gating structures, including forget gates, input gates, and output gates. They are particularly adept at capturing and learning long-range dependencies in time-series data, such as how morning temperatures affect afternoon heat load peaks. The core computation process within a unit of the LTM network is as follows:

[0047] ;

[0048] in, It is the input vector at time step t; and These are the hidden state and the unit state of the previous time step, respectively; These are the activation values ​​of the forget gate, input gate, and output gate, which determine which information is forgotten, which new information is stored, and which information is output. It is a candidate cell state; It refers to the cell state and hidden state after the current time step update; and These are the weight matrix and bias vector that the model needs to learn; It is the sigma activation function, which compresses the value to between 0 and 1, representing the degree to which the gate is open or closed; It is the hyperbolic tangent activation function; This represents element-wise multiplication. Building upon this, an attention mechanism is introduced. This mechanism mimics human cognitive processes, enabling the model to dynamically assign different importance weights to different parts of the input sequence when processing sequential data. In heat load prediction, this means the model can automatically learn that when predicting summer afternoon heat load, irradiance features are more important than wind speed features, while in winter nights, temperature features dominate. The attention weights are calculated as follows:

[0049] ;

[0050] in, Let be the hidden state of the Long Short-Term Memory network at the j-th input time step; It is the state of the prediction model when decoding the t-th output; It is the alignment score, which measures the degree of matching between the input position j and the output position t; These are the weight parameters that need to be learned in the attention mechanism; It is the normalized attention weight, which represents the degree of attention paid to the j-th input information when generating the t-th prediction value; The context vector is the weighted sum of all input hidden states based on attention weights, containing the most useful input information for the current prediction. Historical heat load datasets and historical meteorological data are preprocessed according to model input requirements and then input into this deep neural network model for training. The training objective is to minimize the error between the model's predictions and the actual values, typically using mean squared error as the loss function. After training, the model's weight parameters are fixed, forming the final heat load prediction model. Finally, the generated prediction input feature vector is input into this trained heat load prediction model. The model utilizes its learned complex patterns to process this comprehensive feature vector, which includes future weather, historical load patterns, and the current system state, ultimately outputting a sequence—the heat load demand prediction sequence for the future control period—containing predicted heat load values ​​for multiple future time steps. Figure 3 As shown in the figure, the sequence of gray solid dotted lines represents the actual heat load evolution trajectory under historically similar boundary conditions, while the sequence of black dashed squares represents the future heat load demand prediction sequence generated by the attention mechanism and the long short-term memory network (LSTM) deep neural network structure in this invention.

[0051] For example, the generated 5-dimensional predicted input feature vector [150.3, -25.1, 10.8, -5.2, 1.9] is input into a pre-trained heat load prediction model, which is trained based on the substation's historical heat load and meteorological data from the past two years. When this vector is input into the LSTM network layer as the input for the current time step, the gating units begin to function: the forget gate reduces the reliance on nighttime low-temperature memories based on the component reflecting high irradiance, while the input gate updates the unit state with new daytime feature information. The attention mechanism evaluates the internal state during LSTM processing, calculating different context vectors for predictions of the next 4 hours (10:00 to 13:00). For example, when predicting the heat load at 12:00, the attention mechanism assigns higher weights to the irradiance-related feature components because the model has learned from historical data that midday irradiance is a key factor affecting heat load. The context vector calculated by the attention mechanism is combined with the LSTM hidden state and the predicted value is output hourly through a fully connected output layer: the prediction for the first hour is 128 kW, and then the model generates the predicted values ​​for subsequent time points based on the prediction results of the previous time point and the internal state. The model outputs the complete heat load demand prediction sequence for the future regulation period [128, 145, 152, 135] (unit: kW).

[0052] The dynamic response characteristic parameters of each energy conversion device in the multi-energy complementary system are obtained, and the device dynamic response matrix is ​​constructed based on the dynamic response characteristic parameters;

[0053] Optionally, constructing the device dynamic response matrix based on the dynamic response characteristic parameters includes:

[0054] A step test signal is sent to each energy conversion device, and the response curve of the device's output power changing over time is collected.

[0055] The parameters of the first-order inertial plus pure delay element are identified on the response process curve to obtain the dynamic response time constant and ramp rate limit value of each energy conversion device.

[0056] The dynamic response time constant and the ramp rate limit are combined in a matrix to obtain the device dynamic response matrix.

[0057] Specifically, a step test signal is sent to the device under test (DUT), which is a control signal that instantly switches from a stable power output command value to another command value. For example, the power command of a gas boiler in standby mode is instantly increased from 0 kW to 50% of its rated power. This abrupt signal can fully stimulate the dynamic response process of the device. Simultaneously with sending the step test signal, the actual output power of the device is collected and recorded at high frequency over time, thus forming a response process curve. This curve visually shows the entire process from receiving the command to actually reaching the target power, which is usually not instantaneous but exhibits a certain delay and inertia. Mathematical modeling and analysis are then performed on the collected response process curve, i.e., parameter identification of the first-order inertia plus pure delay element. This model is a classic model widely used in engineering to describe the dynamic response of physical systems, and is crucial for calculating the actual output power of the device over time. ,have:

[0058] ;

[0059] in, The initial stable output power of the equipment before the step test begins; This is the power change amplitude of the step test signal, i.e., the difference between the target power and the initial power; The pure delay time is the "dead zone" time from when the signal is sent until the device's output power begins to change. This is the base of the natural logarithm. By fitting the data of the response process curve to the above formula, two core dynamic response characteristic parameters can be identified. The first parameter is the dynamic response time constant, denoted as... It characterizes how quickly a device overcomes its inertia and approaches a new target power. In the above formula, This is the time constant of the exponential term, and its physical meaning is the time required for the device's output power to change from the initial change to approximately 63.2% of the total change. The smaller the value, the faster the equipment response. The second parameter is the ramp rate limit, which represents the maximum rate at which the equipment's output power can change. It is an inherent upper limit of the equipment's mechanical and thermodynamic performance, and is determined by calculating the maximum slope of the response curve in the initial stage of the response. Theoretically, for a first-order inertial element, the maximum slope occurs at the instant the response begins, and its value is... In practice, the maximum rate of change can be directly extracted from the measured data as the ramp rate limit to reflect the most realistic physical limitations. The above testing and parameter identification process is repeated for all N energy conversion devices to obtain the dynamic response time constant and ramp rate limit for each device. These two dynamic response characteristic parameters of all devices are then combined into an N×2 device dynamic response matrix. This matrix clearly summarizes the dynamic performance of all schedulable heat sources in the entire system.

[0060] For example, in a multi-energy complementary system, there are two energy conversion devices: a gas-fired boiler (device 1) and an electric boiler (device 2). A step test signal is sent to the gas-fired boiler, which is currently operating at a stable output power of 20 kW, instructing it to increase its power to 80 kW. The response curve of its output power over time is collected. For the first 15 seconds after the signal is sent, the boiler's output power remains constant at 20 kW, after which it begins to increase. At 55 seconds after the signal is sent, the output power is recorded as 57.8 kW. Based on this response curve, the parameters of a first-order inertial plus pure delay element are identified. The pure delay time is directly read from the data as 15 seconds. Given: t = 55 seconds. kilowatt, kilowatt, kilowatt, Seconds. Substitute the data for calculation: Solving for The dynamic response time constant of the gas-fired boiler is 40.2 seconds. Analysis of the slope of the response curve revealed that the slope is largest when the power begins to change after t=15 seconds, with a value of 1.5 kW / s. The ramp-up rate limit for the gas-fired boiler is 1.5 kW / s. The same test was then performed on the electric boiler (equipment 2). The electric boiler responded rapidly, and its dynamic response time constant was identified as 5 seconds, with a ramp-up rate limit of 10 kW / s. Combining the dynamic response characteristic parameters of these two devices into a matrix yields the device dynamic response matrix: The first row of this matrix [40.2, 1.5] fully describes the dynamic characteristics of the gas-fired boiler, namely its response time constant of 40.2 seconds and maximum power change rate of 1.5 kW / s. The second row [5, 10] describes the dynamic characteristics of the electric boiler.

[0061] Using the heat load demand forecast sequence as the tracking target and combining it with the equipment dynamic response matrix, multi-objective rolling time-domain optimization is performed to obtain the pre-scheduling plan sequence.

[0062] The pre-scheduling plan sequence obtained by solving includes:

[0063] The actual grid-connected power and the upper limit of available power of renewable energy power generation equipment are collected, and the time-by-time difference is accumulated and a penalty weight that increases stepwise with the duration of abandonment is added to generate a renewable energy abandonment penalty item.

[0064] The dynamic energy consumption characteristics of each energy conversion device during variable load operation are obtained, and the transient energy consumption corresponding to the current output of the device is superimposed with the additional energy consumption generated by start-stop switching to generate an operating energy consumption item.

[0065] The system operating cost function is generated by weighting and adding the renewable energy abandonment penalty term and the operating energy consumption term hourly according to the control cycle.

[0066] The actual and available renewable energy generation for each time period are obtained, and the ratio is calculated for each time period. Then, the total heat demand of the substation for each time period is used as the weight to calculate the average and generate the renewable energy utilization rate target.

[0067] The heat load demand forecast sequence is used as a heat power balance constraint, the equipment dynamic response matrix is ​​used as a equipment output change rate constraint, and upper and lower power limits of each energy conversion equipment are added.

[0068] The system operating cost function and the renewable energy utilization rate target are combined into a single objective function, and a rolling time-domain optimization solution is performed to generate a pre-scheduling plan sequence.

[0069] Specifically, to quantify economic costs, a system operating cost function was constructed. This function consists of two parts. The first part is a renewable energy curtailment penalty term. By collecting real-time data on the actual grid-connected power of renewable energy power generation equipment such as photovoltaics and solar thermal power, and the upper limit of available power during the same period, the unutilized renewable energy power at each moment, i.e., the curtailed power, is calculated. To suppress long-term continuous curtailment, a penalty weight that increases in a stepwise manner with the duration of curtailment is introduced. The curtailment penalty cost at time t is calculated. ,have:

[0070] ;

[0071] in, The duration of continuous abandonment up to time t; Let t be the upper limit of available power from renewable energy sources; The renewable energy power to be used at time t; For a random The penalty weight function uses a stepped-incrementing approach, with a weight of 0.1 for hours in the range [0,1); 0.3 for hours in the range [1,3); 0.6 for hours in the range [3,6); and 1.0 for hours in the range [6 hours and above). This setting aims to prioritize the use of renewable energy and avoid resource waste by increasing economic penalties. The specific step values ​​of this weight function are determined based on statistical analysis of historical abandonment and calculation of renewable energy generation costs. The second part of the system operating cost function is the operating energy consumption item. It includes the energy consumption costs generated by each energy conversion device during operation, as well as the additional costs incurred by start-up and shutdown operations. The total operating energy consumption cost of all devices at calculation time t is calculated as follows. ,have:

[0072] ;

[0073] Where N is the total number of energy conversion devices; It is the output power of device i at time t; Device i has an output power of The cost of fuel or electricity consumption at that time , The nonlinear loss coefficient is... The linear performance consumption coefficient, The no-load or standby loss coefficient is obtained by the least squares fitting method based on the equipment's factory performance curve or on-site variable load test data. It is the additional energy consumption cost of device i per startup, which is obtained based on the measured energy consumption data of the device startup process; This is a binary variable; it is 1 when device i changes from off to on at time t, and 0 otherwise. Adding these two cost components hourly over the entire optimization time domain yields the system operating cost function. ,have:

[0074] ;

[0075] in, The optimization time step is, for example, 1 hour. The goal of this function is to minimize [the time step]. Secondly, to quantify environmental benefits, a renewable energy utilization rate target is constructed. This target aims to maximize the proportion of renewable energy consumption, especially during periods of high heat load demand. [The text then discusses calculating the weighted average renewable energy utilization rate.] ,have:

[0076]

[0077] in, It is the predicted value at time t from the heat load demand forecast sequence; It is a very small positive constant. The goal of this function is to maximize it. For example... Figure 4 As shown, within 12 consecutive scheduling periods, the light gray bars represent the theoretical maximum power generation corresponding to the concurrent available power limit of renewable energy generation equipment in each period, while the dark gray bars represent the actual renewable energy power absorbed and connected to the grid under the rolling time-domain optimization command. After defining the optimization objective, a series of constraints need to be set to ensure the physical feasibility of the plan. First, the thermal power balance constraint: at any time t, the total output power of all heat sources must be equal to the predicted value of the thermal load demand at that time. Second, the equipment output change rate constraint: the power adjustment speed of the equipment cannot exceed its physical limit. This constraint directly uses the ramp rate limit value in the equipment dynamic response matrix, that is:

[0078] ;

[0079] in, Let be the ramp rate limit value for device i, which is the second element in the i-th row of the device's dynamic response matrix. Third, power upper and lower limit constraints for each energy conversion device: the operating power of the device must be within its rated range, i.e.:

[0080] ;

[0081] in, and These are the minimum and maximum technical outputs of device i; This is a binary variable representing the start-up and shutdown status of the equipment. The system operating cost function and the renewable energy utilization rate objective are combined into a single objective function, which is then solved using rolling time-domain optimization. For this single objective function... ,have:

[0082] ;

[0083] in, This is a weighting coefficient used to balance the relationship between economic costs and renewable energy utilization rates. It can be set by the operator based on actual needs such as environmental pressure and economic benefits. This single objective function and all constraints together constitute a mixed-integer nonlinear programming problem. A rolling time-domain optimization approach is used to solve it. That is, at the current moment, the optimization problem over a longer future time domain, such as 24 hours, is solved to obtain a pre-scheduled plan sequence. However, only the plan for the first time step of this sequence is executed. Then, at the next time step, the latest system state and forecast information are obtained, and the above optimization process is repeated. This approach allows the scheduling plan to continuously adapt to the dynamic changes of the system.

[0084] For example, the predicted heat load demand sequence is [128, 145] kW. The dynamic response matrix of the equipment is obtained, where the ramp rate limit for gas-fired boiler (equipment 1) is 50 kW / h, and for electric boiler (equipment 2) it is 200 kW / h. The upper limit of available renewable energy power is [30, 50] kW. A single objective function is constructed: if the weights... Establish constraints, including thermal power balance constraints: Equipment output change rate constraint, at t=0 : Power limits: Gas boiler [20,100] kW, Electric boiler [0,80] kW. The objective function and constraints described above are input into the optimization solver. After calculation, the solver may obtain the following pre-scheduled plan sequence: at time t=1: , , .verify: (Satisfies equilibrium); (Satisfies the climbing requirement). Time t=2: , , .verify: (Satisfies equilibrium). (Meets ramp-up requirements). This result indicates that in the first hour, the gas-fired boiler should output 98 kW, the electric boiler should remain off, and 30 kW of renewable energy should be fully utilized. In the second hour, the gas-fired boiler output should be adjusted to 95 kW, and 50 kW of renewable energy should be fully utilized. This sequence, containing the output values ​​of each device for each time period in the future, is the final generated pre-scheduling plan sequence.

[0085] The thermal delay characteristic parameters of the heat transfer network and the capacity status of the thermal storage device are obtained. The pre-scheduling plan sequence is modified into an execution scheduling instruction that satisfies thermal inertia constraints. Based on the execution scheduling instruction, control signals for each energy conversion device and thermal storage device are generated and sent out for execution.

[0086] Optionally, modifying the pre-scheduled plan sequence into execution scheduling instructions that satisfy thermal inertia constraints includes:

[0087] Based on the thermal delay characteristic parameters, feature extraction is performed to obtain segment length parameters, pipe diameter parameters, and thermal resistance parameters of insulation material, and the pipe thermal delay time coefficient is calculated.

[0088] Based on the capacity status, feature extraction is performed to obtain the current heat storage capacity and rated heat storage and release power value of the thermal storage device. Based on the pipeline thermal delay time coefficient and the rated heat storage and release power value, heat transfer delay constraints and thermal storage power capacity constraints are established.

[0089] The output timing of each device and the thermal storage charging and discharging plan in the pre-scheduled plan sequence are compared and corrected with the thermal transfer delay constraint and the thermal storage power capacity constraint to generate an execution scheduling instruction.

[0090] Specifically, the first step is to obtain the thermal delay characteristic parameters of the heat transfer network. A heat transfer network refers to a piping system connecting heat sources and heat load users. Its thermal delay characteristic parameters include the segment length parameters, pipe diameter parameters, and thermal resistance parameters of the outer insulation material of each pipe constituting the network. These parameters are the fundamental data for network design and construction. Based on these parameters, a key indicator can be calculated: the pipe thermal delay time coefficient, which characterizes the time required for heat to be transferred within the pipe. The calculation of this pipe thermal delay time coefficient... ,have:

[0091] ;

[0092] in, Let be the segment length parameter of the k-th segment of the pipeline network; This reference flow velocity is calculated based on the design flow rate and pipe diameter parameters of the pipeline network, or determined by taking the average operating flow velocity after performing 200 fluid dynamics simulations of the pipeline network under different operating conditions. This formula simplifies the complex fluid thermodynamic process into a concentrated time delay, making it convenient for application in scheduling models. Figure 5As shown, the heat map clearly reveals a logarithmic surface relationship: the thermal delay increases dramatically with longer pipe lengths and lower flow rates. The next step is to obtain the capacity status of the thermal storage device. This includes two key data points read in real time by sensors: the current heat storage capacity of the thermal storage device and the device's rated heat storage / release power. The current heat storage capacity represents the total amount of heat energy currently stored in the storage tank, while the rated heat storage / release power defines the maximum capacity of the device to store or release heat per unit time. Based on the above quantified physical characteristics, two core constraints are established. The first is the heat transfer delay constraint, which describes the causal relationship between heat production and consumption over time. The second is the thermal storage power capacity constraint, which ensures that all operations of the thermal storage device are within its physical limits. The thermal storage power capacity constraint ensures that all charging and discharging operations of the thermal storage equipment are feasible within its physical limits: the thermal storage power and discharging power at any given time are limited to their respective rated maximum values, while the real-time heat storage capacity of the thermal storage tank must always be between its minimum and maximum thermal storage capacity. The dynamic recursive relationship of the heat storage capacity indicates that the heat storage capacity at the end of the current time period is determined by the heat storage capacity at the end of the previous time period plus the net increment of the thermal storage power and discharging power in the current time period after efficiency conversion, multiplied by the time step. This integrates equipment capacity, power limits, and conversion efficiency into the scheduling model, providing a quantified feasible domain boundary for subsequent adjustments and optimizations considering electricity price arbitrage or thermal inertia compensation. The output timing of each equipment and the thermal storage charging and discharging plan in the pre-scheduled plan sequence are compared and corrected with the newly established heat transfer delay constraint and thermal storage power capacity constraint on a time-by-time basis. This correction process is a closed loop of checking and adjustment. For each time step t in the future control time domain, it is first checked whether the thermal storage charging and discharging plan arranged for that time period in the pre-scheduled plan sequence meets the thermal storage power capacity constraint. If the limit is exceeded, it is corrected to the maximum allowable value. Based on the heat transfer delay constraint, the actual heat power arriving at the load side at time t is calculated. This power is the sum of the planned outputs of all heat sources after considering the heat transfer delay time. This arriving power is compared with the predicted heat load demand at time t. If a deviation exists, it is compensated using the available charging and discharging capacity of the thermal storage equipment at time t. If the thermal storage capacity is insufficient to fully compensate for the deviation, the output plan of the heat sources at the historical time corresponding to the heat delay time coefficient of the pipeline needs to be adjusted. This adjustment process prioritizes equipment with fast response speed and low adjustment cost. After time-by-time correction across the entire time domain, a final execution scheduling instruction that fully conforms to physical reality is obtained. Based on this corrected execution scheduling instruction, specific control signals for each energy conversion device and thermal storage device are generated, such as power setpoints and start / stop commands, which are sent to the corresponding equipment controllers through the industrial control network to complete the final scheduling execution.

[0093] For example, the initial pre-scheduling plan covers the next 2 hours with a time step of 1 hour. The total output of the heat source at t=1 and t=2 is 128kW and 145kW respectively, which perfectly matches the predicted heat load demand sequence [128, 145]kW. Due to the 3600-second thermal delay in the heat transfer network, the total length of the network is 3600 meters, and the reference flow velocity is 1 m / s, the heat generated by the heat source at t=1 actually reaches the load side at t=2. Therefore, the heat power balance needs to be rechecked for each time period. At t=1, the actual output of the heat source at t=0 (130kW) needs to meet the 128kW load, resulting in a 2kW surplus. Therefore, the thermal storage equipment is charged at 2kW, increasing the heat storage capacity from 30kWh to 32kWh, which meets the capacity range of 0 to 100kWh and the 40kW rated power constraint. At time t=2, the 128kW output from the heat source at time t=1 is needed to meet the 145kW load, resulting in a 17kW shortage. Therefore, the thermal storage equipment is scheduled to discharge at 17kW, reducing the stored heat to 15kWh, which still satisfies all constraints. The final revised execution instructions are: at time t=1, the gas boiler output is 98kW, renewable energy is 30kW, and thermal storage charging is 2kW; at time t=2, the gas boiler output is 95kW, renewable energy is 50kW, and thermal storage discharge is 17kW. Based on this, corresponding power setting signals are sent to the controllers of the gas boiler and thermal storage equipment.

[0094] Optionally, the method further includes:

[0095] Collect the actual temperature signal at a designated hot node within the substation, and calculate the temperature deviation between the actual temperature signal and the target temperature corresponding to the heat load demand prediction sequence.

[0096] Based on the temperature deviation, feedback compensation calculation is performed to generate a dynamic compensation power correction value;

[0097] The dynamic compensation power correction value is fed forward and superimposed into the next round of execution scheduling instructions.

[0098] Specifically, the process begins by selecting representative locations within the substation as designated thermal nodes, such as temperature-sensitive equipment rooms or main office areas, and installing temperature sensors at these locations. The sensors then collect the actual temperature signals of these nodes in real time, obtaining continuous or discrete time-series data. Simultaneously, a target temperature signal is needed for comparison with the actual temperature signal. The predicted heat load demand sequence needs to be transformed into a corresponding target temperature sequence. This transformation is based on a simplified model describing the thermodynamic characteristics of the building or space, which correlates the heat power required to maintain a specific target temperature. A simplified linear relationship can be expressed as:

[0099] ;

[0100] in, It is the target temperature at time t; It is the predicted value at time t in the obtained heat load demand forecast sequence; It is a reference temperature reached by the environment when there is no heating input, such as the temperature when the temperature difference between indoors and outdoors is zero; This is a heat-to-temperature conversion coefficient, representing the temperature increase achieved per unit of heating power. It is obtained through thermal calibration of substation buildings or statistical analysis of historical operating data, such as linear regression analysis of 300 data points with different heating powers and corresponding stable room temperatures. After acquiring the actual temperature signal and calculating the target temperature, the temperature deviation between the two can be calculated. Based on the calculated temperature deviation, feedback compensation calculations are performed to generate a dynamic compensation power correction value for revising future heating plans. This calculation process employs a controller that comprehensively considers the current deviation, historical cumulative deviation, and deviation trend. The dynamic compensation power correction value calculated at time t is... ,have:

[0101] ;

[0102] in, The gain is proportional, making the correction power proportional to the current deviation, enabling a fast response to deviations; This is the integral gain, used to eliminate steady-state biases caused by model inaccuracies or persistent disturbances; The differential gain is used to predictively adjust the system based on the rate of change of the deviation, thereby suppressing system oscillations and improving stability. The settings of these three gain parameters are determined by engineering tuning methods such as the step response method after testing the dynamic response characteristics of the controlled thermodynamic system. Let be the temperature deviation at time t. The calculated dynamic compensation power correction value is fed forward and superimposed into the execution scheduling instruction of the next round. This means that the dynamic compensation power correction value is not used to adjust the equipment output at the current moment, but is added as a correction term to the objective function or constraint of the next round of rolling optimization. In the next round of optimization calculation, the thermal power balance constraint will be corrected to:

[0103] ;

[0104] in, It is calculated in the previous scheduling cycle and corresponds to a future time. The dynamic compensation power correction value. In this way, the optimization process will take into account and compensate for known system deviations in advance, generating a scheduling plan that is closer to the actual needs.

[0105] For example, the current time is 10:00 AM, and scheduling instructions are being executed to prepare for the next scheduling round starting at 11:00 AM. In the resulting heat load demand forecast sequence, the forecast value for 10:00 AM is 128 kW. The target temperature at 10:00 AM is: At 10:00, the actual temperature signal at the designated hot spot within the substation was collected, and the reading was 23.9°C. The temperature deviation was calculated. Feedback compensation calculations are performed based on this temperature deviation. The controller parameters have been tuned as follows: The calculation period is 1 minute. At 9:59, the temperature deviation is 0.4°C, and the cumulative error up to 9:59 is 3.0°C·min. The dynamic compensation power correction value is calculated at 10:00. When optimization is performed after 10:00 for the 11:00 timeframe and beyond, the heat power balance constraints will change. For example, for the 11:00 timeframe, if the original heat load demand forecast is 145 kW, the revised balance objective will become: The optimization algorithm will use 150.78 kW as the heating target at 11:00 to solve the pre-scheduling plan sequence, thus incorporating the compensation for the temperature deviation at 10:00 into the future scheduling plan in a feedforward manner.

[0106] Optionally, the method further includes:

[0107] Obtain real-time electricity price signals and ancillary service demand signals from the electricity market interface;

[0108] The acquired real-time electricity price signal is incorporated into the system operating cost function to calculate the electricity purchase cost or electricity sales revenue. At the same time, an ancillary service revenue item is added based on the ancillary service demand signal, and the system operating cost function is updated.

[0109] Based on the capacity status and the real-time electricity price signal, a thermal storage charge-discharge arbitrage constraint adapted to electricity price fluctuations is generated.

[0110] Based on the updated system operating cost function and the thermal storage charge-discharge arbitrage constraint, a rolling time-domain optimization solution is performed to obtain a combined heat and power pre-schedule plan sequence coordinated with electricity market signals.

[0111] Specifically, the process begins by acquiring two key external signals from the electricity market interface. The first is the real-time electricity price signal, a time series predicting the electricity trading price for each scheduling period (e.g., hourly) over a future period, such as the next 24 hours. The second is the ancillary service demand signal, issued by the grid operator to maintain grid stability, such as requiring the system to provide a certain amount of reserve capacity or participate in frequency regulation, with corresponding compensation prices. After acquiring these signals, the system operating cost function is first updated. The original cost function mainly focused on internal operating costs and renewable energy consumption; now, the interaction costs and benefits with the electricity market need to be incorporated. The updated system operating cost function is achieved by adding two new terms. The first is the cost of purchasing or selling electricity, calculated based on the real-time electricity price signal and the power interaction between the system and the grid. The second is the ancillary service revenue term, calculated based on the ancillary service demand signal and the planned service volume provided by the system. The updated system operating cost function... ,have:

[0112] ;

[0113] in, and These are the original renewable energy abandonment penalty items and equipment operation energy consumption items; It is the real-time electricity price signal at time t obtained from the electricity market interface; It is the interaction power between the power grid and the grid at time t, and it is the optimization variable. When When electricity is purchased from the grid, the cost of purchasing electricity is incurred; when This indicates that electricity is sold to the power grid, generating revenue from electricity sales; It is the ancillary service compensation price at time t determined based on the ancillary service demand signal; This refers to the ancillary services, such as reserve capacity, that the system plans to provide at time t. By minimizing this updated system operating cost function, the optimization algorithm automatically finds the optimal strategy for purchasing and selling electricity and providing ancillary services. To fully utilize the flexibility of thermal storage devices for economic arbitrage, a thermal storage charge-discharge arbitrage constraint adapted to electricity price fluctuations needs to be established. This constraint, based on the current capacity status of the thermal storage devices and real-time electricity price signals, guides the thermal storage devices to utilize cheap electricity to store heat during low-price periods (off-peak hours) and release the stored heat during high-price periods (peak hours) to replace high-cost heat sources, thereby reducing the total cost. The thermal storage charge-discharge arbitrage constraint can be transformed into logical rules by setting low-price and high-price thresholds: when the real-time electricity price is lower than or equal to the low-price threshold, heat storage operation is prioritized, i.e., heat release power is limited to zero; when the real-time electricity price is higher than or equal to the high-price threshold, heat release operation is prioritized, i.e., heat storage power is limited to zero. The low-price and high-price thresholds are set based on statistical analysis of historical electricity price data, typically using the 25th and 75th percentiles of the past year's real-time electricity price as reference values. Based on the above, a rolling time-domain optimization solution is performed again. The resulting optimization is a combined thermal and power pre-dispatch plan sequence coordinated with electricity market signals. This sequence includes not only the output plans of each energy conversion device, but also the power purchase and sale plan with the grid, the ancillary service provision plan, and the thermal storage charging and discharging plan closely coupled with electricity price signals.

[0114] For example, real-time electricity prices (0.3 yuan / kWh during off-peak hours and 0.9 yuan / kWh during peak hours) and ancillary service demand (10kW of upward reserve provided during time period t=2, with a compensation of 0.1 yuan / kWh) are obtained from the electricity market interface and incorporated into the system operating cost function. The predicted heat load demand sequence is [128, 145] kW, including gas boilers, electric boilers, and thermal storage equipment. A low electricity price threshold of 0.4 yuan / kWh and a high electricity price threshold of 0.8 yuan / kWh are set to form a price-driven charge-discharge arbitrage constraint: the low-price thermal storage logic is triggered during off-peak hours t=1, i.e., the thermal release power is 0; the high-price thermal release logic is triggered during peak hours t=2, i.e., the thermal storage power is 0. Based on this, the optimizer makes the following decisions: During off-peak hours, the marginal cost of the electric boiler (0.3 yuan / kWh) is lower than that of the gas boiler (0.5 yuan / kWh), so the output of the electric boiler is increased, with a total heating capacity of 138kW. After meeting the 128kW load, the remaining 10kW is used for thermal storage charging. During peak hours, the cost of the electric boiler rises to 0.9 yuan / kWh, so the optimizer chooses to shut down the electric boiler. The 145kW load is met by the gas boiler (85kW), renewable energy (50kW), and thermal storage discharge (10kW). At the same time, the gas boiler has a 15kW reserve capacity and can commit to providing 10kW of ancillary services to generate revenue. The final generated combined heat and power pre-dispatch plan is as follows: t=1: Gas boiler 40kW, electric boiler 68kW, renewable energy 30kW, thermal storage charging 10kW; t=2: Gas boiler 85kW, electric boiler shut down, renewable energy 50kW, thermal storage discharge 10kW, ancillary services 10kW. Compared to solutions that do not consider market signals, this plan achieves electricity price arbitrage through off-peak energy storage and peak-peak energy release, and participates in the ancillary services market, thereby reducing operating costs and increasing revenue.

[0115] Based on the same inventive concept, such as Figure 6 As shown, the present invention also provides a multi-energy complementary thermal management system for substations, the system comprising:

[0116] The feature fusion module is used to acquire meteorological and environmental forecast data of the substation location, historical operating data of heat load in the substation, and real-time energy status data of the multi-energy complementary system. It performs time-frequency domain feature extraction on the meteorological and environmental forecast data and the historical operating data, and performs feature fusion with the real-time energy status data to generate a prediction input feature vector.

[0117] The load forecasting module is used to input the forecasting input feature vector into a pre-built heat load forecasting model to obtain a heat load demand forecast sequence for future control periods.

[0118] The response matrix module is used to obtain the dynamic response characteristic parameters of each energy conversion device in the multi-energy complementary system, and to construct the device dynamic response matrix based on the dynamic response characteristic parameters.

[0119] The rolling optimization module is used to perform multi-objective rolling time-domain optimization by taking the heat load demand prediction sequence as the tracking target and combining it with the equipment dynamic response matrix to obtain the pre-scheduling plan sequence.

[0120] The inertia correction module is used to obtain the thermal delay characteristic parameters of the heat transfer network and the capacity status of the heat storage device, correct the pre-scheduling plan sequence into an execution scheduling instruction that satisfies thermal inertia constraints, generate control signals for each energy conversion device and heat storage device based on the execution scheduling instruction, and issue them for execution.

[0121] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

Claims

1. A method for coordinated control of multi-energy complementary thermal management in substations, characterized in that, The method includes: Acquire meteorological and environmental forecast data of the substation location, historical operating data of heat load within the substation, and real-time energy status data of the multi-energy complementary system. Extract time-frequency domain features from the meteorological and environmental forecast data and the historical operating data, and fuse them with the real-time energy status data to generate a predictive input feature vector. The predicted input feature vector is input into a pre-built heat load prediction model to obtain a heat load demand prediction sequence for future control periods. The dynamic response characteristic parameters of each energy conversion device in the multi-energy complementary system are obtained, and the device dynamic response matrix is ​​constructed based on the dynamic response characteristic parameters; Using the heat load demand forecast sequence as the tracking target and combining it with the equipment dynamic response matrix, multi-objective rolling time-domain optimization is performed to obtain the pre-scheduling plan sequence. The thermal delay characteristic parameters of the heat transfer network and the capacity status of the thermal storage device are obtained. The pre-scheduling plan sequence is modified into an execution scheduling instruction that satisfies thermal inertia constraints. Based on the execution scheduling instruction, control signals for each energy conversion device and thermal storage device are generated and sent out for execution.

2. The substation multi-energy complementary thermal management and coordinated control method according to claim 1, characterized in that, The generated prediction input feature vector includes: Obtain the time series of temperature, irradiance, and wind speed from the meteorological environment forecast data, as well as the heat load power series corresponding to the time series of temperature, irradiance, and wind speed from the historical operating data; Wavelet transform is performed on the time series of temperature, irradiance, and wind speed, and the heat load power series to extract multi-scale time-frequency feature components. The multi-scale time-frequency feature components and the real-time energy state data are concatenated and their dimensions reduced to generate a predictive input feature vector.

3. The substation multi-energy complementary thermal management and coordinated control method according to claim 1, characterized in that, The obtained heat load demand forecast sequence for the future regulation period includes: Obtain the heat load dataset of the substation for historical periods and the historical meteorological environment data corresponding to the historical periods; Construct a deep neural network model structure using attention mechanisms and long short-term memory networks; The historical heat load dataset and the historical meteorological environment data are input into the deep neural network model structure for training to obtain a heat load prediction model. Combined with the prediction input feature vector, a heat load demand prediction sequence is output.

4. The substation multi-energy complementary thermal management coordinated control method according to claim 1, characterized in that, The construction of the device dynamic response matrix based on the dynamic response characteristic parameters includes: A step test signal is sent to each energy conversion device, and the response curve of the device's output power changing over time is collected. The parameters of the first-order inertial plus pure delay element are identified on the response process curve to obtain the dynamic response time constant and ramp rate limit value of each energy conversion device. The dynamic response time constant and the ramp rate limit are combined in a matrix to obtain the device dynamic response matrix.

5. The method for coordinated control of multi-energy complementary thermal management in a substation according to claim 1, characterized in that, The pre-scheduling plan sequence obtained by solving includes: Construct a system operating cost function that includes a renewable energy curtailment penalty and the energy consumption of each energy conversion device, as well as a renewable energy utilization rate target; The heat load demand forecast sequence is used as a heat power balance constraint, the equipment dynamic response matrix is ​​used as a equipment output change rate constraint, and upper and lower power limits of each energy conversion equipment are added. The system operating cost function and the renewable energy utilization rate target are combined into a single objective function, and a rolling time-domain optimization solution is performed to generate a pre-scheduling plan sequence.

6. The substation multi-energy complementary thermal management coordinated control method according to claim 5, characterized in that, The construction of the system operating cost function, which includes a renewable energy curtailment penalty and the energy consumption of each energy conversion device, and the renewable energy utilization target, includes: The actual grid-connected power and the upper limit of available power of renewable energy power generation equipment are collected, and the time-by-time difference is accumulated and a penalty weight that increases stepwise with the duration of abandonment is added to generate a renewable energy abandonment penalty item. The dynamic energy consumption characteristics of each energy conversion device during variable load operation are obtained, and the transient energy consumption corresponding to the current output of the device is superimposed with the additional energy consumption generated by start-stop switching to generate an operating energy consumption item. The system operating cost function is generated by weighting and adding the renewable energy abandonment penalty term and the operating energy consumption term hourly according to the control cycle. The actual and available renewable energy generation for each time period are obtained, and the ratio is calculated for each time period. Then, the total heat demand of the substation for each time period is used as the weight to calculate the average, and the renewable energy utilization rate target is generated.

7. The substation multi-energy complementary thermal management coordinated control method according to claim 1, characterized in that, The step of modifying the pre-scheduled plan sequence to satisfy the thermal inertia constraint includes the following execution scheduling instructions: Based on the thermal delay characteristic parameters, feature extraction is performed to obtain segment length parameters, pipe diameter parameters, and thermal resistance parameters of insulation material, and the pipe thermal delay time coefficient is calculated. Based on the capacity status, feature extraction is performed to obtain the current heat storage capacity and rated heat storage and release power value of the thermal storage device. Based on the pipeline thermal delay time coefficient and the rated heat storage and release power value, heat transfer delay constraints and thermal storage power capacity constraints are established. The output timing of each device and the thermal storage charging and discharging plan in the pre-scheduled plan sequence are compared and corrected with the thermal transfer delay constraint and the thermal storage power capacity constraint to generate an execution scheduling instruction.

8. The substation multi-energy complementary thermal management coordinated control method according to claim 7, characterized in that, The method further includes: Collect the actual temperature signal at a designated hot node within the substation, and calculate the temperature deviation between the actual temperature signal and the target temperature corresponding to the heat load demand prediction sequence. Based on the temperature deviation, feedback compensation calculation is performed to generate a dynamic compensation power correction value; The dynamic compensation power correction value is fed forward and superimposed into the next round of execution scheduling instructions.

9. A method for coordinated control of multi-energy complementary thermal management in a substation according to claim 5, characterized in that, The method further includes: Obtain real-time electricity price signals and ancillary service demand signals from the electricity market interface; The acquired real-time electricity price signal is incorporated into the system operating cost function to calculate the electricity purchase cost or electricity sales revenue. At the same time, an ancillary service revenue item is added based on the ancillary service demand signal, and the system operating cost function is updated. Based on the capacity status and the real-time electricity price signal, a thermal storage charge-discharge arbitrage constraint adapted to electricity price fluctuations is generated. Based on the updated system operating cost function and the thermal storage charge-discharge arbitrage constraint, a rolling time-domain optimization solution is performed to obtain a combined heat and power pre-schedule plan sequence coordinated with electricity market signals.

10. A multi-energy complementary thermal management system for substations, applied to the multi-energy complementary thermal management coordinated control method for substations as described in any one of claims 1-9, characterized in that, The system includes: The feature fusion module is used to acquire meteorological and environmental forecast data of the substation location, historical operating data of heat load in the substation, and real-time energy status data of the multi-energy complementary system. It performs time-frequency domain feature extraction on the meteorological and environmental forecast data and the historical operating data, and performs feature fusion with the real-time energy status data to generate a prediction input feature vector. The load forecasting module is used to input the forecasting input feature vector into a pre-built heat load forecasting model to obtain a heat load demand forecast sequence for future control periods. The response matrix module is used to obtain the dynamic response characteristic parameters of each energy conversion device in the multi-energy complementary system, and to construct the device dynamic response matrix based on the dynamic response characteristic parameters. The rolling optimization module is used to perform multi-objective rolling time-domain optimization by taking the heat load demand prediction sequence as the tracking target and combining it with the equipment dynamic response matrix to obtain the pre-scheduling plan sequence. The inertia correction module is used to obtain the thermal delay characteristic parameters of the heat transfer network and the capacity status of the heat storage device, correct the pre-scheduling plan sequence into an execution scheduling instruction that satisfies thermal inertia constraints, generate control signals for each energy conversion device and heat storage device based on the execution scheduling instruction, and issue them for execution.

Images