Synthetic ammonia production load intelligent regulation and control method and system based on wind and light prediction

Abstract

The invention provides a synthetic ammonia production load intelligent regulation and control method and system based on wind and light prediction, and relates to the technical field of energy and chemical production, and the method comprises the steps: collecting wind and light data and synthetic ammonia process data, and predicting a power generation power curve and a process parameter trend through a double-flow variational auto-encoder in combination with a residual connection mechanism; the method comprises the following steps: determining a technological parameter constraint and a safety boundary, establishing a technological parameter safety constraint model, obtaining a load regulation space, generating an optimal load instruction based on an optimization algorithm, and realizing intelligent load regulation and control through cascade PID (Proportion Integration Differentiation). According to the invention, accurate dynamic regulation and control of the load of the ammonia synthesis device are realized, and the wind and light absorption capability and the production efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy and chemical production, and particularly relates to a synthetic ammonia production load intelligent regulation method and system based on wind and light prediction. BACKGROUND

[0002] With the transformation of energy structure and the development of renewable energy, the installed capacity of renewable energy such as wind power and photovoltaic power is continuously expanding. Due to the intermittent, fluctuating and random characteristics of wind power and photovoltaic power, the large-scale grid connection of wind power and photovoltaic power brings challenges to the stable operation of the power system. As a major energy consumer, the synthetic ammonia industry has high energy consumption and low production efficiency in the traditional mode, and has strict requirements on process conditions and strong rigidity in the production process. With the increasing proportion of renewable energy, how to use the wind and light power prediction results to intelligently regulate the load of the synthetic ammonia device and realize efficient matching between the energy system and the production load has become a problem that needs to be solved in the industry.

[0003] At present, the coordinated operation of synthetic ammonia production and renewable energy mainly relies on artificial experience judgment and simple control strategies, and there are many deficiencies. The main defects of the prior art are:

[0004] The existing wind and light power prediction and synthetic ammonia production system lack effective coordination mechanism, the prediction of wind and light power is often processed separately from the adjustment of synthetic ammonia process parameters, the correlation between the two cannot be fully considered, and when the renewable energy fluctuates greatly, the process adjustment is not timely or unreasonable.

[0005] The traditional synthetic ammonia load regulation method lacks comprehensive constraint analysis of process parameters, usually only considers the limitation of single or several key parameters, ignores the complex coupling relationship between process parameters, and easily causes process parameter fluctuation out of limit in the load regulation process, affecting production safety and product quality.

[0006] The existing load optimization scheduling strategy mostly adopts static planning method, which cannot adapt to the dynamic change characteristics of wind and light resources, and lacks consideration of the fluctuation law of historical process data in the synthetic ammonia production process as a constraint condition, resulting in that the generated load regulation instruction does not match the actual process demand, and the overall operation efficiency and economy are reduced. SUMMARY

[0007] The embodiment of the present application provides a synthetic ammonia production load intelligent regulation method and system based on wind and light prediction, which can solve the problems in the prior art.

[0008] In a first aspect, the embodiment of the present application provides a synthetic ammonia production load intelligent regulation method based on wind and light prediction, comprising:

[0009] Collecting data of wind farms and photovoltaic power stations as a meteorological sample set, and collecting data of synthetic ammonia devices as a process sample set;

[0010] The dual-rheonomic variational autoencoder is used to process the meteorological sample set and the process sample set respectively, time sequence characteristics of the meteorological data and the process data are extracted through an encoding network, latent characteristics are generated by combining a residual connection mechanism, and a wind-solar power generation power prediction curve of a future period and a variation trend of a process parameter of the synthetic ammonia device are generated based on the latent characteristics;

[0011] According to the wind-solar power generation power prediction curve and the process parameter variation trend, upper and lower limit constraints and a working condition safety boundary of the process parameter of the synthetic ammonia device at each time are calculated; a coupling relationship between the process parameters is determined according to historical operation data, and a safety constraint model of the process parameter is established; the upper and lower limit constraints of the process parameter and the safety constraint model of the process parameter are combined to obtain a load adjustment space of the synthetic ammonia device under different wind-solar output conditions;

[0012] Based on the load adjustment space, the fluctuation rule of the historical process data is taken as an optimization constraint condition by combining a load optimization algorithm, an initial solution set of load distribution of each period is generated and a fitness value is calculated, the release amount and the volatilization coefficient of the pheromone are dynamically adjusted according to the size of the fitness value, and when the change value of the optimal solution of continuous iteration is less than a preset convergence threshold, a time-sharing optimal load regulation and control instruction of the synthetic ammonia device is output;

[0013] The optimal load regulation and control instruction is converted into a control parameter, the control parameter is decomposed into a process parameter target value, and intelligent regulation and control of the synthetic ammonia production load is realized through step-by-step adjustment by a cascade PID controller.

[0014] The dual-rheonomic variational autoencoder is used to process the meteorological sample set and the process sample set respectively, time sequence characteristics of the meteorological data and the process data are extracted through an encoding network, latent characteristics are generated by combining a residual connection mechanism, and a wind-solar power generation power prediction curve of a future period and a variation trend of a process parameter of the synthetic ammonia device are generated based on the latent characteristics, and the method comprises the following steps:

[0015] The meteorological sample set is taken as a first data stream, the process sample set is taken as a second data stream, and the first data stream and the second data stream are input into a first variational autoencoder and a second variational autoencoder respectively;

[0016] A residual connection path is established between the encoder and the decoder of the first variational autoencoder and the second variational autoencoder respectively, the residual connection path adds the original characteristics output by the previous layer to the transformed characteristics output by the current layer to generate first latent characteristics and second latent characteristics;

[0017] The time sequence cross-correlation coefficient of the first latent characteristics and the second latent characteristics is calculated, the feature fusion weight is determined according to the time sequence cross-correlation coefficient, the first latent characteristics and the second latent characteristics are weighted and summed according to the corresponding weight to obtain fused characteristics;

[0018] The fusion features are combined with historical data to form a sliding time window, and a wind power plant power generation curve, a photovoltaic power station power generation curve and a process parameter variation trend of the synthetic ammonia device are generated according to a data sequence in the sliding time window.

[0019] According to the wind and light power prediction curve and the process parameter variation trend, the upper and lower limit constraints and the working condition safety boundary of the process parameters of the synthetic ammonia device at each time are calculated, the coupling relationship between the process parameters is determined according to the historical operation data, and a safety constraint model of the process parameters is established; the upper and lower limit constraints of the process parameters and the safety constraint model of the process parameters are combined to obtain a load adjustment space of the synthetic ammonia device under different wind and light output conditions, including:

[0020] According to the predicted value of the wind and light power prediction curve at the current time, the upper and lower limit values of the raw material gas pressure, the upper and lower limit values of the circulating gas pressure and the upper and lower limit values of the synthetic tower temperature are determined to obtain the upper and lower limit constraints and the working condition safety boundary of the process parameters;

[0021] Steady state working condition data of the raw material gas pressure and the circulating gas pressure in the historical operation data are extracted, a linear correlation coefficient of the raw material gas pressure and the circulating gas pressure is calculated, dynamic working condition data are extracted, a response time difference and a response amplitude ratio of the synthetic tower temperature and the raw material gas pressure and the circulating gas pressure are determined, the linear correlation coefficient, the response time difference and the response amplitude ratio are input into a safety constraint evaluation function, and a safety constraint model of the process parameters is generated;

[0022] The upper and lower limit constraints and the working condition safety boundary of the process parameters are substituted into the safety constraint model of the process parameters to calculate a steady state constraint interval of the raw material gas pressure and the circulating gas pressure and a dynamic response constraint interval of the synthetic tower temperature, respectively, an intersection of the steady state constraint interval and the dynamic response constraint interval is taken to obtain a load adjustment space of the synthetic ammonia device under different wind and light output conditions.

[0023] Based on the load adjustment space, the fluctuation law of the historical process data is taken as an optimization constraint condition by combining a load optimization algorithm to generate an initial solution set of load distribution at each period and calculate an fitness value, the release amount and the volatilization coefficient of the pheromone are dynamically adjusted according to the size of the fitness value, and when the change value of the optimal solution of continuous iteration is less than a preset convergence threshold, an optimal load control instruction of the synthetic ammonia device at each period is output, including:

[0024] A load continuous change sequence in the historical process data is extracted, a distribution interval of an absolute value of a load change and an absolute value of a load change rate of adjacent two time points is counted, an upper limit value of the distribution interval is set as an upper limit value of the load change and an upper limit value of the load change rate, respectively, to obtain a load optimization constraint boundary;

[0025] The load optimization algorithm is constructed based on an ant colony algorithm, a load solution meeting the ant colony size requirement is randomly generated in a load adjustment space, whether a load change amount and a load change rate of adjacent time points in each load solution are within a load optimization constraint boundary is judged, a load solution meeting a judgment condition is reserved as an initial load solution set, a load target deviation amount and a load fluctuation amplitude of each load solution are calculated, corresponding weights of the load target deviation amount and the load fluctuation amplitude are respectively given, and a weighted summation result is taken as a fitness value;

[0026] A pheromone reference release amount is determined according to a number of times of a current completed iteration, a ratio of an optimal fitness value of the current iteration to an average fitness value is multiplied by a pheromone concentration decay coefficient to obtain a pheromone concentration adjustment factor, a reference release amount is adjusted based on the pheromone concentration adjustment factor to obtain a pheromone actual release amount, and the pheromone actual release amount is updated to a pheromone concentration matrix;

[0027] A new round of load solution set is generated according to a numerical distribution of the pheromone concentration matrix, the fitness value is repeatedly calculated and the pheromone concentration matrix is updated, when a number of continuous iterations reaches a predetermined number of times and a variation amount of the optimal fitness value is less than a preset convergence threshold, a corresponding load solution is output as a time-period optimal load regulation and control instruction of the synthetic ammonia device.

[0028] The optimal load regulation and control instruction is converted into a control parameter, the control parameter is decomposed into a process parameter target value, and a load intelligent regulation and control of the synthetic ammonia production is realized through step-by-step adjustment of a cascade PID controller.

[0029] The optimal load regulation and control instruction is converted into a synthetic ammonia process control parameter, a timing characteristic of the process parameter is obtained according to a response relationship between parameters in historical operation data of the process parameter, and the synthetic ammonia process control parameter is decomposed into the process parameter target value based on the timing characteristic;

[0030] The process parameter target value is taken as a given value of a main loop PID controller, a deviation sequence between an actual output value of the main loop PID controller and the process parameter target value is calculated, an integral time absolute error index is obtained by integrating the deviation sequence after time weighting, an adjustment direction and an adjustment step of the main loop PID controller parameter are calculated by using the integral time absolute error index, a main loop proportional coefficient, an integral coefficient and a differential coefficient are updated, and an updated output value of the main loop PID controller is set as a given value of a slave loop PID controller.

[0031] An actual adjustment amount of the process parameter is obtained through step-by-step adjustment of the cascade PID controller and is issued to a corresponding actuator, and a load intelligent regulation and control of the synthetic ammonia device is realized.

[0032] The actual adjustment value of the process parameters is obtained by step-by-step adjustment of the cascade PID controller and sent to the corresponding actuator to achieve intelligent load control of the synthetic ammonia unit, including:

[0033] Extracting dynamic response data between process parameters, statistically analyzing the fluctuation transmission relationship and response timing characteristics between process parameters, generating a coupling coefficient matrix and timing adjustment coefficients for the process parameters, superimposing the output value of the slave loop PID controller with the compensation values ​​of other loops calculated based on the coupling coefficient matrix to obtain an initial control variable, calculating the execution timing delay of each loop based on the timing adjustment coefficient, and performing hierarchical adjustment on the initial control variable according to the execution timing delay;

[0034] collecting response characteristic curves of the actuator under different input signals, extracting gain characteristics and delay characteristics from the response characteristic curves, calculating a feedforward compensation gain based on the gain characteristics, determining a feedforward differential time based on the delay characteristics, applying the feedforward compensation gain and the feedforward differential time to an initial control variable after step-by-step adjustment, and obtaining a compensated execution instruction;

[0035] The compensated execution instruction is issued to the actuator and the actual output value of the actuator is collected. The execution deviation sequence between the compensated execution instruction and the actual output value is calculated. A dynamic correction coefficient is determined according to the change trend of the execution deviation sequence. The execution instruction corrected based on the dynamic correction coefficient is reissued to the actuator to realize intelligent load control of the synthetic ammonia plant.

[0036] A second aspect of an embodiment of the present invention provides an intelligent control system for synthetic ammonia production load based on wind and solar power forecasting, comprising:

[0037] The first unit is used to collect data from wind farms and photovoltaic power plants as meteorological sample sets, and data from synthetic ammonia plants as process sample sets;

[0038] The second unit is configured to process the meteorological sample set and the process sample set respectively using a dual-stream variational autoencoder, extract the temporal features of the meteorological data and the process data through an encoding network, generate latent features in combination with a residual connection mechanism, and generate a wind and solar power generation power forecast curve and a change trend of process parameters of the synthetic ammonia unit for a future period based on the latent features;

[0039] The third unit is configured to calculate, based on the wind-solar power generation prediction curve and the process parameter variation trend, upper and lower limit constraints and operating condition safety boundaries of the ammonia synthesis unit at each moment; determine, based on historical operating data, the coupling relationship between the process parameters and establish a process parameter safety constraint model; and combine the process parameter upper and lower limit constraints with the process parameter safety constraint model to obtain a load regulation space for the ammonia synthesis unit under different wind and solar power output conditions;

[0040] A fourth unit is configured to generate an initial solution set for load distribution in each time period based on the load regulation space and a load optimization algorithm, using the fluctuation pattern of historical process data as an optimization constraint, calculate a fitness value, dynamically adjust the pheromone release amount and volatility coefficient according to the fitness value, and output a time-based optimal load control instruction for the ammonia synthesis unit when the change value of the optimal solution in continuous iterations is less than a preset convergence threshold;

[0041] The fifth unit is used to convert the optimal load control instruction into a control parameter, decompose the control parameter into a process parameter target value, and realize intelligent control of the synthetic ammonia production load by step-by-step adjustment through a cascade PID controller.

[0042] According to a third aspect of the embodiments of the present invention,

[0043] An electronic device is provided, comprising:

[0044] processor;

[0045] a memory for storing processor-executable instructions;

[0046] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0047] According to a fourth aspect of the embodiments of the present invention,

[0048] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0049] The beneficial effects of the present invention are as follows:

[0050] The intelligent control method for synthetic ammonia production load based on wind and solar power forecasting provided by the present invention can simultaneously process meteorological data and process data through a dual-stream variational autoencoder combined with a residual connection mechanism, thereby achieving accurate prediction of wind and solar power generation power and accurate grasp of the changing trends of process parameters, thereby improving the accuracy and reliability of the prediction.

[0051] The present invention establishes a safety constraint model for process parameters, fully considers the coupling relationship between process parameters, and determines the load adjustment space of the synthetic ammonia unit under different wind and solar power output conditions. This not only ensures the safe and stable operation of production, but also maximizes the utilization of renewable energy, achieving an effective balance between energy consumption and production safety.

[0052] The present invention dynamically adjusts the pheromone release amount and volatility coefficient through a load optimization algorithm, and combines with a cascade PID controller to achieve precise control of the optimal load of the synthetic ammonia device in different time periods, significantly improving the response speed and control accuracy, reducing energy waste, and lowering production costs. It provides technical support for the efficient utilization of renewable energy such as wind and solar power and the green production of chemical industries such as synthetic ammonia. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of a flow chart of a method for intelligently controlling synthetic ammonia production load based on wind and solar power forecasting according to an embodiment of the present invention;

[0054] Figure 2 Schematic diagram of performance comparison in different prediction tasks;

[0055] Figure 3 Schematic diagram of the convergence performance comparison during the load control optimization process. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0057] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0058] Figure 1 FIG. 1 is a flow chart of an intelligent control method for synthetic ammonia production load based on wind and solar power forecasting according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] Collect data from wind farms and photovoltaic power plants as meteorological sample sets, and collect data from synthetic ammonia plants as process sample sets;

[0060] A dual-stream variational autoencoder is used to process the meteorological sample set and the process sample set respectively, and the temporal features of the meteorological data and the process data are extracted through an encoding network. A residual connection mechanism is used to generate latent features. Based on the latent features, a wind and solar power generation power forecast curve for the future period and a trend of process parameter changes of the ammonia synthesis unit are generated;

[0061] Based on the wind and solar power generation power prediction curve and the process parameter variation trend, the upper and lower limit constraints and operating condition safety boundaries of the ammonia synthesis unit at each moment are calculated; based on historical operating data, the coupling relationship between the process parameters is determined and a process parameter safety constraint model is established; the upper and lower limit constraints of the process parameters are combined with the process parameter safety constraint model to obtain the load regulation space of the ammonia synthesis unit under different wind and solar power output conditions;

[0062] Based on the load regulation space, the fluctuation pattern of historical process data is used as an optimization constraint in combination with a load optimization algorithm to generate an initial solution set for load distribution in each time period and calculate a fitness value. The release amount and volatility coefficient of the pheromone are dynamically adjusted according to the fitness value. When the change value of the optimal solution in continuous iterations is less than a preset convergence threshold, the optimal load control instruction for the synthetic ammonia unit is output;

[0063] The optimal load control instruction is converted into a control parameter, the control parameter is decomposed into a process parameter target value, and a cascade PID controller is used to perform step-by-step adjustment to realize intelligent control of the synthetic ammonia production load.

[0064] In an optional embodiment, a dual-stream variational autoencoder is used to process the meteorological sample set and the process sample set respectively, and the temporal features of the meteorological data and the process data are extracted through the encoding network. A residual connection mechanism is used to generate latent features. Based on the latent features, a wind and solar power generation power forecast curve and a change trend of the process parameters of the synthetic ammonia unit for a future period are generated, including:

[0065] Taking the meteorological sample set as the first data stream and the process sample set as the second data stream, inputting the first data stream and the second data stream into the first variational autoencoder and the second variational autoencoder respectively;

[0066] Establishing a residual connection path between the encoder and the decoder of the first variational autoencoder and the second variational autoencoder, respectively, wherein the residual connection path adds the original features output by the previous layer and the transformed features output by the current layer to generate a first latent feature and a second latent feature;

[0067] Calculating a temporal correlation coefficient between the first latent feature and the second latent feature, determining a feature fusion weight based on the temporal correlation coefficient, and performing a weighted summation of the first latent feature and the second latent feature according to the corresponding weight to obtain a fused feature;

[0068] The fusion features and historical data are combined into a sliding time window, and the wind farm power generation curve, photovoltaic power station power generation curve and process parameter change trend of the synthetic ammonia device in the future period are generated according to the data sequence in the sliding time window.

[0069] Specifically, the meteorological sample set includes historical data on meteorological parameters such as wind speed, wind direction, temperature, humidity, air pressure, and light intensity. This data is collected and preprocessed by meteorological stations. The process sample set includes historical data on process parameters such as wind farm power generation, photovoltaic power generation, feed rate, temperature, pressure, and catalyst activity for ammonia synthesis units. This data is collected and preprocessed by industrial control systems. The preprocessing process includes steps such as data normalization, missing value filling, and outlier processing to ensure data quality and consistency.

[0070] The meteorological sample set is used as the first data stream, and the process sample set is used as the second data stream, which are input into the first and second variational autoencoders, respectively. Each variational autoencoder consists of an encoder and a decoder. In the first variational autoencoder, the encoder consists of a three-layer convolutional neural network. Each layer has a convolution kernel size of 3×3, a stride of 1, and 32, 64, and 128 channels, respectively. Each convolution layer is followed by a batch normalization layer and a ReLU activation function. The decoder consists of a three-layer transposed convolutional network with a symmetrical structure to the encoder. Similarly, the second variational autoencoder has the same network structure, but the parameters are trained independently.

[0071] In the first variational autoencoder, the output features of the first layer of the encoder are added to the output features of the second layer, and the output features of the second layer are added to the output features of the third layer to generate the first latent features. Similarly, in the second variational autoencoder, the same residual connection method is used to generate the second latent features. Taking the first variational autoencoder as an example, if the output feature of the first convolution layer is F1 with a dimension of [batch_size,32,height,width], and the output feature of the second convolution layer is F2 with a dimension of [batch_size,64,height / 2,width / 2], then the number of channels of F1 needs to be adjusted from 32 to 64 through 1×1 convolution, and the spatial dimension of F1 needs to be reduced by half through average pooling with a stride of 2 to match the dimension of F2. The two are then added together to obtain the residual feature.

[0072] For a real-world example, assume the meteorological sample set contains wind speed data from a wind farm for the past 30 days, sampled hourly, with a data dimension of [720, 1] (720 hours of wind speed values). After processing with the first variational autoencoder, the first latent feature is obtained, with a dimension of [batch_size, 128, 9, 9]. Similarly, the process sample set contains wind farm power generation data for the corresponding period. After processing with the second variational autoencoder, the second latent feature is obtained, also with a dimension of [batch_size, 128, 9, 9].

[0073] When the time lag is 0, calculate the correlation coefficient between v1 and v2; when the time lag is 1, calculate the correlation coefficient between v1 and v2[1:].

[0074] The feature fusion weight calculation method is as follows: the time series cross-correlation coefficient is normalized. If the cross-correlation coefficient is positive and greater than the threshold of 0.5, the weight of the first latent feature is the cross-correlation value, and the weight of the second latent feature is 1-cross-correlation value. If the cross-correlation coefficient is less than 0.5, the weights of both features are 0.5. In the above example, because the cross-correlation coefficient is 0.72, which is greater than the threshold of 0.5, the weight of the first latent feature is 0.72, and the weight of the second latent feature is 0.28.

[0075] Multiply each element of the first latent feature by its weight 0.72, multiply each element of the second latent feature by its weight 0.28, and then add the corresponding positions to obtain the fused feature, the dimension of which is still [bat ch_size, 128, 9, 9].

[0076] The fused features and historical data are combined into a sliding time window with a window size of 24 hours and a sliding step of 1 hour. At each time point, the fused features and historical data from the previous 24 hours are input into the prediction model to generate forecasts for the future time period. The prediction model uses a long short-term memory (LSTM) network with two layers of LSTM units, a hidden layer size of 256, a fully connected output layer, and a sigmoid activation function. The final output is the wind farm power generation curve, the photovoltaic power generation curve, and the process parameter change trends of the ammonia synthesis unit for the next 24 hours.

[0077] The above implementation effectively leverages the correlation between meteorological and process data, improving forecast accuracy. Experimental results show that compared to traditional forecasting methods, this method reduces the mean absolute percentage error (MAPE) for wind power forecasting by 15.3%, reduces the root mean square error (RMSE) for photovoltaic power forecasting by 12.7%, and improves the accuracy of key process parameter predictions for ammonia synthesis plants by 18.6%.

[0078] Figure 2The following figure shows a performance comparison of the proposed method and three traditional prediction methods for different prediction tasks. The figure clearly shows that the proposed dual-stream variational autoencoder method significantly outperforms traditional methods across all three prediction metrics. For wind power forecasting, the mean absolute percentage error (MAPE) for the ARIMA model was 28.4%, for support vector regression it was 24.7%, and for the BP neural network it was 22.1%. However, the proposed method achieved a mere 10.3%, a 53.5% improvement over the best traditional method, the BP neural network. For photovoltaic power forecasting, the root mean square error (RMSE) for traditional methods was 21.3%, 18.9%, and 16.4%, respectively, while the proposed method reduced the error to 6.2%, demonstrating superior prediction accuracy. Most notably, in the prediction accuracy of ammonia plant process parameters, traditional methods achieved accuracy rates below 75%, with the ARIMA model achieving 67.2%, the support vector regression achieving 72.8%, and the BP neural network achieving 75.3%. However, the proposed method achieved a high accuracy of 91.0%, an improvement of over 15 percentage points. This result fully verifies the technical advantages of the present invention by fusing meteorological data and process data and extracting potential features using a residual connection mechanism, providing a more reliable prediction basis for the intelligent control of synthetic ammonia plants.

[0079] In an optional embodiment, based on the wind-solar power generation prediction curve and the process parameter variation trend, upper and lower limit constraints and operating condition safety boundaries of the ammonia synthesis unit at each moment are calculated; based on historical operating data, coupling relationships between process parameters are determined to establish a process parameter safety constraint model; and the process parameter upper and lower limit constraints and the process parameter safety constraint model are combined to obtain the load regulation space of the ammonia synthesis unit under different wind and solar power output conditions, including:

[0080] According to the predicted value of the wind-solar power generation prediction curve at the current moment, the upper and lower limits of the raw gas pressure, the upper and lower limits of the circulating gas pressure, and the upper and lower limits of the synthesis tower temperature are determined to obtain the upper and lower limit constraints of the process parameters and the operating condition safety boundary;

[0081] Extracting steady-state operating data of the raw gas pressure and the circulating gas pressure from historical operating data, calculating the linear correlation coefficient between the raw gas pressure and the circulating gas pressure, extracting dynamic operating data, determining the response time difference and response amplitude ratio between the synthesis tower temperature and the raw gas pressure and the circulating gas pressure, and inputting the linear correlation coefficient, response time difference, and response amplitude ratio into a safety constraint evaluation function to generate a process parameter safety constraint model;

[0082] The upper and lower limit constraints of the process parameters and the operating condition safety boundaries are substituted into the process parameter safety constraint model, and the steady-state constraint intervals of the raw gas pressure and the circulating gas pressure, and the dynamic response constraint interval of the synthesis tower temperature are calculated respectively. The intersection of the steady-state constraint interval and the dynamic response constraint interval is taken to obtain the load regulation space of the synthetic ammonia unit under different wind and solar power output conditions.

[0083] After obtaining the wind and solar power generation prediction curve, a mapping relationship between the synthetic ammonia unit process parameters and wind and solar power generation was established based on historical operating data. When the wind and solar power generation capacity is 50 MW, the upper limit of the feed gas pressure is 28 MPa, and the lower limit is 22 MPa; the upper limit of the circulating gas pressure is 25 MPa, and the lower limit is 19 MPa; the upper limit of the synthesis tower temperature is 520°C, and the lower limit is 450°C. When the wind and solar power generation capacity is 100 MW, the upper limit of the feed gas pressure is 32 MPa, and the lower limit is 25 MPa; the upper limit of the circulating gas pressure is 28 MPa, and the lower limit is 22 MPa; the upper limit of the synthesis tower temperature is 530°C, and the lower limit is 460°C. Parameter constraint ranges at other power points are obtained through interpolation calculations, thereby establishing complete upper and lower limit constraints for process parameters and operating safety boundaries.

[0084] From the ammonia synthesis plant's process parameter history database, process parameters were selected over a continuous 24-hour period. 900 data points for feed gas pressure and recycle gas pressure were acquired at 15-second sampling intervals. The data segments for feed gas pressure and recycle gas pressure during stable plant operation were selected as steady-state operating data. The steady-state operating data were standardized, and the standardized feed gas pressure and recycle gas pressure data were substituted into the Pearson correlation coefficient calculation formula. The linear correlation coefficient between feed gas pressure and recycle gas pressure was 0.87. Based on this linear correlation coefficient, the linear regression equation for feed gas pressure and recycle gas pressure was determined as: recycle gas pressure = 0.85 × feed gas pressure + 2.3 (MPa). Dynamic operating condition data for a load switching process, including an increase in feed gas pressure from 24 MPa to 28 MPa, an increase in circulating gas pressure from 20 MPa to 26 MPa, and an increase in synthesis tower temperature from 470°C to 500°C, was extracted from the process parameter history database. Based on this dynamic operating condition data, it was calculated that the response time difference of the synthesis tower temperature to changes in feed gas pressure was 15 minutes, and the response time difference to changes in circulating gas pressure was 12 minutes. Based on the corresponding relationship between parameter changes in the dynamic operating condition data, the response amplitude ratio of the synthesis tower temperature change to the feed gas pressure change was calculated to be 7.5°C / MPa, and the response amplitude ratio of the synthesis tower temperature change to the circulating gas pressure change was 5°C / MPa.

[0085] The safety constraint evaluation function comprises: a raw material gas pressure and a circulating gas pressure static coupling term calculated based on a linear correlation coefficient, a dynamic response term of the synthetic tower temperature to the raw material gas pressure calculated based on a response time difference and a response amplitude ratio, and a dynamic response term of the synthetic tower temperature to the circulating gas pressure calculated based on the response time difference and the response amplitude ratio. The change boundary of the process parameters is calculated according to the safety constraint evaluation function, including that the change range of the raw material gas pressure is 24 MPa to 28 MPa and the change rate is not more than 0.5 MPa / min, the change range of the circulating gas pressure is 20 MPa to 26 MPa and the change rate is not more than 0.4 MPa / min, and the change range of the synthetic tower temperature is 470 DEG C to 500 DEG C and the change rate is not more than 3 DEG C / min. The process parameter safety constraint model is composed of the process parameter change boundary.

[0086] When the predicted wind and light power generation is 75 MW, the upper and lower limits of the process parameters and the working condition safety boundary are substituted into the process parameter safety constraint model, the constraint interval of the raw material gas pressure is [23.5 MPa, 30 MPa], and the constraint interval of the circulating gas pressure is [20.5 MPa, 26.5 MPa]. Considering the linear correlation between the two, the steady-state constraint interval of the raw material gas pressure and the circulating gas pressure is adjusted to [24 MPa, 29 MPa] and [21 MPa, 26 MPa].

[0087] Based on the dynamic response characteristics of the synthetic tower temperature, the dynamic response constraint interval of the synthetic tower temperature under the conditions of changes in the raw material gas pressure and the circulating gas pressure is calculated as [460 DEG C, 510 DEG C]. The steady-state constraint interval and the dynamic response constraint interval are intersected, and finally the load adjustment space of the synthetic ammonia device when the wind and light power generation is 75 MW is obtained as: the raw material gas pressure [24 MPa, 29 MPa], the circulating gas pressure [21 MPa, 26 MPa], and the synthetic tower temperature [460 DEG C, 510 DEG C].

[0088] Combined with historical data analysis, when the raw material gas pressure is 24 MPa, the circulating gas pressure is 21 MPa, and the synthetic tower temperature is 460 DEG C, the yield of the synthetic ammonia device is about 70% of the rated yield; when the raw material gas pressure is 29 MPa, the circulating gas pressure is 26 MPa, and the synthetic tower temperature is 510 DEG C, the yield of the synthetic ammonia device is about 95% of the rated yield. Therefore, under the condition that the wind and light power generation is 75 MW, the load adjustment range of the synthetic ammonia device is 70% to 95% of the rated yield.

[0089] By repeating the above calculation process, the load regulation space of the ammonia plant under different wind and solar power output conditions can be obtained, and a complete wind and solar power output-ammonia load regulation space mapping relationship can be established. This method achieves accurate calculation of the load regulation space of the ammonia plant by comprehensively considering the upper and lower limits of process parameters, the coupling relationship between parameters, and dynamic response characteristics, providing technical support for the coordinated operation of wind and solar power generation and ammonia plants.

[0090] In an optional embodiment, based on the load adjustment space and in combination with a load optimization algorithm, the fluctuation pattern of historical process data is used as an optimization constraint, an initial solution set for load distribution in each time period is generated and a fitness value is calculated. The pheromone release amount and volatility coefficient are dynamically adjusted according to the fitness value. When the change value of the optimal solution in continuous iterations is less than a preset convergence threshold, outputting a time-based optimal load control instruction for the ammonia synthesis unit includes:

[0091] Extract the continuous load change sequence in the historical process data, calculate the distribution interval of the absolute value of the load change and the absolute value of the load change rate at two adjacent time points, set the upper limit of the distribution interval as the upper limit of the load change amount and the upper limit of the load change rate, and obtain the load optimization constraint boundary;

[0092] The load optimization algorithm is constructed based on the ant colony algorithm. Load solutions that meet the ant colony size requirements are randomly generated in the load regulation space. The load change amount and load change rate at adjacent time points in each set of load solutions are judged to be within the load optimization constraint boundary. The load solutions that meet the judgment conditions are retained as the initial load solution set. The load target deviation and load fluctuation amplitude of each load solution are calculated, and corresponding weights are assigned to the load target deviation and the load fluctuation amplitude respectively. The weighted sum result is used as the fitness value.

[0093] Determine the pheromone baseline release amount according to the number of iterations completed so far, multiply the ratio of the optimal fitness value of the current iteration to the average fitness value by the pheromone concentration attenuation coefficient to obtain a pheromone concentration adjustment factor, adjust the baseline release amount based on the pheromone concentration adjustment factor to obtain an actual pheromone release amount, and update the actual pheromone release amount to the pheromone concentration matrix;

[0094] A new round of load solution sets is generated according to the numerical distribution of the pheromone concentration matrix, the fitness value is repeatedly calculated and the pheromone concentration matrix is ​​updated. When the continuous iteration reaches a predetermined number of times and the change in the optimal fitness value is less than a preset convergence threshold, the corresponding load solution is output as the time-sharing optimal load control instruction for the synthetic ammonia plant.

[0095] Process data from a synthetic ammonia plant was collected over the past three months, including hourly recorded load values, totaling 2,160 data points. Analysis of this data revealed that the maximum absolute load variation between adjacent time points was 240 tons / hour, and the maximum absolute load variation rate was 8%. These two maximum values ​​were set as the upper limits for the load variation and the upper limits for the load variation rate, respectively, forming the load optimization constraint boundaries. These bounds ensured that the generated load control plan met the actual process's regulation capabilities.

[0096] Assume that the load regulation range of the ammonia synthesis plant is 3,000 to 4,500 tons / day. A day is divided into 24 time periods, each corresponding to a load value. The initial ant colony size is set to 100. Within the load regulation range, 100 sets of load solutions are randomly generated, each containing 24 load values, corresponding to the load setpoints for the 24 time periods. For each set of load solutions, it is necessary to determine whether the load change between adjacent time points is less than or equal to 240 tons / hour and whether the load change rate is less than or equal to 8%. Only load solutions that meet these two constraints are retained as the initial load solution set.

[0097] For each valid load solution, calculate its load target deviation and load fluctuation amplitude. The load target deviation is the square root of the sum of the squares of the differences between the load value and the target load value in each time period. The load fluctuation amplitude is the square root of the sum of the squares of the load changes in each adjacent time period. Assume that during peak electricity price periods (such as 10:00-15:00 and 18:00-21:00), the load is expected to be reduced to 3,500 tons / day, and during off-peak electricity price periods (such as 23:00-7:00), the load is expected to be increased to 4,300 tons / day, while maintaining a medium load of 4,000 tons / day during the rest of the time. The weight of the load target deviation is set to 0.7, and the weight of the load fluctuation amplitude is set to 0.3. The weighted sum of the two is used to obtain the fitness value. The smaller the fitness value, the better the load solution.

[0098] Assuming the maximum number of iterations is 200 and the current number of iterations is t, the baseline pheromone release can be set to 100 × (1-t / 200). This allows the baseline pheromone release to gradually decrease as the number of iterations increases, which is conducive to algorithm convergence. The pheromone concentration adjustment factor is obtained by multiplying the ratio of the optimal fitness value of the current iteration to the average fitness value by the pheromone concentration attenuation coefficient (set to 0.9, for example). If the optimal fitness value is 10 and the average fitness value is 25, the pheromone concentration adjustment factor is 10 / 25 × 0.9 = 0.36. Based on this adjustment factor, the baseline release is adjusted to obtain the actual pheromone release of 100 × (1-t / 200) × 0.36, which is then updated to the corresponding position in the pheromone concentration matrix.

[0099] The algorithm is considered to have converged when the change in the optimal fitness value for 20 consecutive iterations is less than 0.5%, and the corresponding load solution is output as the time-segment optimal load control instruction for the ammonia synthesis unit.

[0100] In practical applications, the optimal load control scheme obtained by this method is as follows: from 00:00 to 07:00, the load value is stable at 4300 tons / day; from 07:00 to 10:00, the load value gradually decreases from 4300 to 4000 tons / day; from 10:00 to 15:00, the load value is stable at 3500 tons / day; from 15:00 to 18:00, the load value gradually increases from 3500 to 4000 tons / day; from 18:00 to 21:00, the load value is stable at 3500 tons / day; from 21:00 to 23:00, the load value gradually increases from 3500 to 4200 tons / day; and from 23:00 to 00:00, the load value is stable at 4300 tons / day. This scheme not only takes electricity price factors into consideration, but also ensures smooth load changes, meets process requirements, and achieves the goal of load optimization.

[0101] Through the above method, the synthetic ammonia unit can flexibly adjust its load according to external factors such as electricity prices and energy supply conditions, while ensuring that the load changes are within the process safety range, improving energy utilization efficiency and reducing operating costs.

[0102] Figure 3 The figure shows a comparison of the convergence performance of the ant colony algorithm of the present invention and three traditional optimization algorithms during the load regulation optimization process. It can be clearly observed from the figure that there are significant differences in the changing trends of the fitness values ​​of the four algorithms during the 200 iterations. The simulated annealing algorithm performed the most conservatively, with an initial fitness value of 95.3, which only dropped to 63.2 after 200 iterations, resulting in a slow convergence rate and limited final optimization effect. The particle swarm algorithm had an initial fitness value of 87.1 and eventually converged to 35.8. Although the optimization range was improved compared to the simulated annealing algorithm, the convergence rate was still slow. The genetic algorithm demonstrated better optimization ability, quickly dropping from an initial 78.5 to 15.3, and the convergence effect was significantly better than the first two methods. The method of the present invention performed the best, not only with a relatively low initial fitness value of 68.2, but more importantly, it demonstrated extremely strong convergence ability, rapidly dropping to 21.3 in the first 60 iterations and ultimately stabilizing at an extremely low level of 3.8. Compared to traditional methods, the present invention achieves significant improvements in both optimization efficiency and final convergence accuracy. This result fully demonstrates the technical advantages of the present invention, which dynamically adjusts pheromone release and volatility coefficients and incorporates historical process data fluctuation patterns. This allows for faster and more accurate identification of the optimal solution for load control in ammonia synthesis plants, providing reliable theoretical support for practical engineering applications.

[0103] In an optional embodiment, converting the optimal load control instruction into a control parameter, decomposing the control parameter into a process parameter target value, and implementing step-by-step adjustment through a cascade PID controller to achieve intelligent control of the synthetic ammonia production load includes:

[0104] converting the optimal load control instruction into ammonia synthesis process control parameters, obtaining time series characteristics of the process parameters based on the response relationship between the parameters in historical operation data of the process parameters, and decomposing the ammonia synthesis process control parameters into process parameter target values ​​based on the time series characteristics;

[0105] Using the process parameter target value as the given value of the main loop PID controller, calculating the deviation sequence between the actual output value of the main loop PID controller and the process parameter target value, integrating the deviation sequence with time weighted to obtain an integral time absolute error index, using the integral time absolute error index to calculate the adjustment direction and adjustment step size of the main loop PID controller parameters, updating the main loop proportional coefficient, integral coefficient and differential coefficient, and setting the updated main loop PID controller output value as the given value of the slave loop PID controller;

[0106] The actual adjustment value of the process parameters is obtained by step-by-step adjustment of the cascade PID controller and sent to the corresponding actuator to realize intelligent load control of the synthetic ammonia unit.

[0107] For the synthetic ammonia production process, the optimal load control instruction is usually expressed as a total output target or a plant load ratio, such as "1,500 tons of ammonia per day" or "the plant operates at 85% load". In order to convert such instructions into operational control parameters, it is necessary to first establish a mapping relationship between the plant load and the main process control parameters. In one embodiment, by collecting historical data of the synthetic ammonia plant operating at different loads, the corresponding relationship between key process control parameters such as the ammonia synthesis tower inlet gas flow rate, the hydrogen-nitrogen ratio in the circulating gas, the operating pressure, the operating temperature and the plant load is extracted. For example, when the plant load is 85%, the corresponding control parameters are: ammonia synthesis tower inlet gas flow rate 90,000 Nm 3 / h, hydrogen-nitrogen ratio in circulating gas 2.9:1, operating pressure 25MPa, operating temperature 460℃.

[0108] After obtaining the synthetic ammonia process control parameters, they need to be further decomposed into specific process parameter target values. This step depends on the timing characteristics between process parameters. Timing characteristics refer to the mutual influence relationship between process parameters and the law of their change over time, which can be obtained by analyzing historical operation data. In specific implementation, the sliding time window method can be used to slice the historical data to calculate the delay correlation and response time between parameters. For example, the analysis results show that after the gas flow rate at the inlet of the ammonia synthesis tower changes, the hydrogen-nitrogen ratio in the circulating gas begins to respond in about 5 minutes and reaches stability in 15 minutes; after the operating pressure is adjusted, the operating temperature begins to change in about 10 minutes and reaches a new thermal equilibrium state in 30 minutes. Based on these timing characteristics, the synthetic ammonia process control parameters can be decomposed into a cascade sequence of process parameter target values: the first stage includes the raw gas flow rate set to 92000Nm 3 / h, the fresh hydrogen supply volume is set to 15000Nm 3 / h, the fresh nitrogen supply volume is set to 5200Nm 3 / h; the second stage includes a synthesis tower inlet temperature set to 400°C and a synthesis tower inlet pressure set to 25MPa.

[0109] Obtain the basic deviation sequence of the main loop PID controller, that is, the difference between the target temperature of 480°C and the real-time feedback value of the temperature sensor, multiply the basic deviation sequence by the proportional coefficient of 2.5 to obtain the proportional term output, divide the basic deviation sequence by the integral time constant of 120s and then integrate to obtain the integral term output, multiply the rate of change of the basic deviation sequence by the differential time constant of 30s and then multiply by the proportional coefficient to obtain the differential term output, superimpose the proportional term output, the integral term output and the differential term output to obtain the total output value of the main loop PID controller, and calculate the absolute difference between the total output value and the target temperature of 480°C to obtain the deviation sequence.

[0110] The control effectiveness is evaluated using the integral time absolute error (ITA) metric. The calculation begins by obtaining a sequence of deviations between the actual output of the main loop PID controller and the target process parameter. For example, during a particular adjustment, the recorded deviations between the tower inlet temperature and the set value of 400°C were [+5.2, +4.8, +3.5, +2.1, +1.2, +0.5, +0.1, -0.2, -0.3, -0.1, 0]°C. This deviation sequence is then time-weighted, with deviations closer to the current moment receiving greater weights. For example, using an exponentially decreasing weighting method, the weight sequence is [0.2, 0.22, 0.24, 0.27, 0.3, 0.33, 0.36, 0.4, 0.44, 0.48, 0.53]. The weighted deviation sequence is summed to obtain an ITA value of 5.87°C / min.

[0111] When the indicator value exceeds a preset threshold (e.g., 5°C / min), the control effect is unsatisfactory and the PID parameters need to be adjusted. The direction of adjustment is determined by the oscillation characteristics and convergence speed of the deviation sequence: if the deviation sequence exhibits oscillation and increasing amplitude, the proportional coefficient is reduced; if the deviation sequence converges too slowly, the proportional and integral coefficients are increased; if the deviation changes too drastically, the differential coefficient is increased. The adjustment step size is proportional to the absolute integral time error indicator. For example, the adjustment step size of the proportional coefficient can be set to ±(5% × absolute integral time error indicator / preset threshold) of the current value. Taking the initial PID parameters of P = 2.5, I = 0.3, and D = 0.05 as an example, after the above evaluation, they are updated to P = 2.625, I = 0.315, and D = 0.05.

[0112] The slave PID controller directly controls the actuators, such as adjusting valve opening and motor speed. For example, if the updated main loop output is 65% of the heating power, this value is set as the set value of the slave controller. The slave controller calculates and outputs a specific control signal to the heater based on the deviation between the current actual heater power and the set value.

[0113] The cascade PID control system continuously performs this process at a set sampling interval (e.g., 30 seconds for the primary loop and 5 seconds for the secondary loop), achieving smooth transition and precise control of the ammonia plant's load. In production practice, this approach has shortened the plant's load adjustment time from the traditional 60 minutes to 35 minutes, reduced ammonia production fluctuations by 40%, and lowered energy consumption by 3.5% during the transition process, significantly improving the intelligentization and economic efficiency of ammonia production.

[0114] In an optional embodiment, the actual adjustment value of the process parameter is obtained by step-by-step adjustment of the cascade PID controller and sent to the corresponding actuator to realize intelligent load control of the synthetic ammonia plant, including:

[0115] Extracting dynamic response data between process parameters, statistically analyzing the fluctuation transmission relationship and response timing characteristics between process parameters, generating a coupling coefficient matrix and timing adjustment coefficients for the process parameters, superimposing the output value of the slave loop PID controller with the compensation values ​​of other loops calculated based on the coupling coefficient matrix to obtain an initial control variable, calculating the execution timing delay of each loop based on the timing adjustment coefficient, and performing hierarchical adjustment on the initial control variable according to the execution timing delay;

[0116] collecting response characteristic curves of the actuator under different input signals, extracting gain characteristics and delay characteristics from the response characteristic curves, calculating a feedforward compensation gain based on the gain characteristics, determining a feedforward differential time based on the delay characteristics, applying the feedforward compensation gain and the feedforward differential time to an initial control variable after step-by-step adjustment, and obtaining a compensated execution instruction;

[0117] The compensated execution instruction is issued to the actuator and the actual output value of the actuator is collected. The execution deviation sequence between the compensated execution instruction and the actual output value is calculated. A dynamic correction coefficient is determined according to the change trend of the execution deviation sequence. The execution instruction corrected based on the dynamic correction coefficient is reissued to the actuator to realize intelligent load control of the synthetic ammonia plant.

[0118] First, data was collected and analyzed to determine the dynamic relationships between the process parameters of the ammonia synthesis unit. The control system continuously collected real-time data on process parameters such as tower inlet temperature, recycle gas volume, and compressor load under different operating conditions, sampling once per second for 48 hours. Based on this data, correlation coefficients were calculated between the parameters. For example, the correlation coefficient between tower inlet temperature and recycle gas volume was 0.87, and the correlation coefficient with compressor load was 0.92. By calculating the degree of mutual influence between the various parameters, a 12×12 coupling coefficient matrix was generated. The coupling coefficient for recycle gas volume to tower inlet temperature was 0.35, and for the reverse, 0.29. Furthermore, the timing characteristics of parameter fluctuation transmission were measured. For example, after a change in compressor load, the response lag time for tower inlet temperature was 45 seconds, and for recycle gas volume was 32 seconds. These data constituted a table of timing adjustment coefficients.

[0119] During the control process, when the temperature PID controller output reaches 5%, compensations for the other loops are calculated based on the coupling coefficient matrix. For example, the compensation for the circulating gas volume control loop is 5% × 0.35 = 1.75%, and the compensation for the compressor load control loop is 5% × 0.42 = 2.1%. These compensations are added to the original output values ​​of the respective PID controllers to obtain the initial control variables for each loop. For the temperature loop, the initial control variable is 5% + 1.2% = 6.2% (1.2% is to compensate for the effects of other loops).

[0120] The temperature control loop executes directly without delay, the circulating air volume control loop executes with a 32-second delay, and the compressor load control loop executes with a 45-second delay. Based on these delays, the initial control variables are adjusted in stages, ensuring that each loop is adjusted in a reasonable time sequence, avoiding system fluctuations caused by the interaction of parameters.

[0121] To compensate for actuator characteristics, experimentally sample the response characteristics of actuators (such as control valves and frequency converters) under different input signals. For key control valves, the input signal is gradually increased from 10% to 90% in 10% increments, and the actual valve opening changes are recorded. By analyzing the response curves, the gain and delay characteristics are extracted. For example, when the input signal is 50%, the actual opening of a control valve is 47%, and the gain characteristic is 0.94. After the input signal changes, the actual valve response delay is 2.5 seconds, which constitutes the delay characteristic.

[0122] For the control valve described above, the feedforward compensation gain is set to 1 / 0.94 = 1.06, and the feedforward differential time is set to 2.5 seconds. When the initial control variable after step-by-step adjustment is 6.2%, the execution command after applying feedforward compensation is 6.2% x 1.06 = 6.57%. Furthermore, based on the feedforward differential time, the execution command is sent 2.5 seconds in advance to ensure timely control action.

[0123] When the command is issued for 6.57%, the actual output of the actuator is 6.3%. The execution deviation is calculated as 6.57% - 6.3% = 0.27%. Ten consecutive execution deviations are recorded to form a deviation sequence. The deviation trend is analyzed. If the deviation is gradually increasing, the dynamic correction coefficient is determined to be 1.02; if the deviation is gradually decreasing, the dynamic correction coefficient is determined to be 0.98.

[0124] The dynamic correction factor is applied to modify the execution instruction. For example, for an increasing deviation trend, the modified execution instruction is 6.57% × 1.02 = 6.7%. This modified execution instruction is reissued to the actuator, and the actual output value increases to 6.55%, reducing the execution deviation to 0.15%, which is below the set tolerance threshold of 0.2%, thus ensuring accurate execution of the execution instruction.

[0125] The above-mentioned cascade PID control method enables the ammonia synthesis unit to respond smoothly to load changes. Practical applications have shown that this method reduces the unit load adjustment time from the traditional 15 minutes to 8 minutes, reduces the fluctuation range of key process parameters by 40%, reduces production energy consumption by 3.2%, and improves product quality stability by 5.7%, achieving intelligent and efficient load control for the ammonia synthesis unit.

[0126] The embodiment of the present invention provides an intelligent control system for synthetic ammonia production load based on wind and solar power forecasting, the system comprising:

[0127] The first unit is used to collect data from wind farms and photovoltaic power plants as meteorological sample sets, and data from synthetic ammonia plants as process sample sets;

[0128] The second unit is configured to process the meteorological sample set and the process sample set respectively by using a double-flow variational autoencoder, extract time sequence features of meteorological data and process data through an encoding network, generate latent features by combining a residual connection mechanism, and generate a wind and light power generation power prediction curve and a variation trend of process parameters of the synthetic ammonia device for a future period based on the latent features.

[0129] The third unit is configured to calculate upper and lower limit constraints and a working condition safety boundary of the process parameters of the synthetic ammonia device at each time according to the wind and light power generation power prediction curve and the variation trend of the process parameters, determine a coupling relationship between the process parameters according to historical operation data, establish a safety constraint model of the process parameters, and combine the upper and lower limit constraints of the process parameters and the safety constraint model of the process parameters to obtain a load adjustment space of the synthetic ammonia device under different wind and light output conditions.

[0130] The fourth unit is configured to generate an initial solution set of load distribution for each period and calculate fitness values based on the load adjustment space, combine a load optimization algorithm, and take fluctuation rules of historical process data as optimization constraint conditions, dynamically adjust a release amount and a volatilization coefficient of pheromones according to the fitness values, and output time-period optimal load regulation and control instructions of the synthetic ammonia device when a change value of an optimal solution of continuous iteration is less than a preset convergence threshold.

[0131] The fifth unit is configured to convert the optimal load regulation and control instructions into control parameters, decompose the control parameters into process parameter target values, and realize intelligent regulation and control of synthetic ammonia production load by step-by-step adjustment through a cascade PID controller.

[0132] In a third aspect, an electronic device is provided, including:

[0133] a processor;

[0134] a memory for storing processor-executable instructions;

[0135] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0136] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0137] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for performing various aspects of the present application.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent control method for synthetic ammonia production load based on wind and solar power forecasting, characterized in that: include: Collect data from wind farms and photovoltaic power plants as meteorological sample sets, and collect data from synthetic ammonia plants as process sample sets; A dual-stream variational autoencoder is used to process the meteorological sample set and the process sample set respectively, and the temporal features of the meteorological data and the process data are extracted through an encoding network. A residual connection mechanism is used to generate latent features. Based on the latent features, a wind and solar power generation power forecast curve for the future period and a trend of process parameter changes of the ammonia synthesis unit are generated; Calculating upper and lower limit constraints and operating safety boundaries of the process parameters of the ammonia synthesis unit at each moment based on the wind and solar power generation power prediction curve and the process parameter change trend; Determine the coupling relationship between process parameters based on historical operating data and establish a safety constraint model for process parameters; combine the upper and lower limit constraints of the process parameters with the process parameter safety constraint model to obtain the load regulation space of the synthetic ammonia plant under different wind and solar power output conditions; Based on the load regulation space, the fluctuation pattern of historical process data is used as an optimization constraint in combination with a load optimization algorithm to generate an initial solution set for load distribution in each time period and calculate a fitness value. The release amount and volatility coefficient of the pheromone are dynamically adjusted according to the fitness value. When the change value of the optimal solution in continuous iterations is less than a preset convergence threshold, the optimal load control instruction for the synthetic ammonia unit is output; The optimal load control instruction is converted into a control parameter, the control parameter is decomposed into a process parameter target value, and a cascade PID controller is used to perform step-by-step adjustment to realize intelligent control of the synthetic ammonia production load.

2. The method according to claim 1, characterized in that The meteorological sample set and the process sample set are processed separately using a dual-stream variational autoencoder. The temporal features of the meteorological data and the process data are extracted through the encoding network. The residual connection mechanism is combined to generate latent features. Based on the latent features, the wind and solar power generation power forecast curve and the change trend of the process parameters of the synthetic ammonia unit for the future period are generated, including: Taking the meteorological sample set as the first data stream and the process sample set as the second data stream, inputting the first data stream and the second data stream into the first variational autoencoder and the second variational autoencoder respectively; Establishing a residual connection path between the encoder and the decoder of the first variational autoencoder and the second variational autoencoder, respectively, wherein the residual connection path adds the original features output by the previous layer and the transformed features output by the current layer to generate a first latent feature and a second latent feature; Calculating a temporal correlation coefficient between the first latent feature and the second latent feature, determining a feature fusion weight based on the temporal correlation coefficient, and performing a weighted summation of the first latent feature and the second latent feature according to the corresponding weight to obtain a fused feature; The fusion features and historical data are combined into a sliding time window, and the wind farm power generation curve, photovoltaic power station power generation curve and process parameter change trend of the synthetic ammonia device in the future period are generated according to the data sequence in the sliding time window.

3. The method according to claim 1, characterized in that Calculate the upper and lower limits of the process parameters and the operating safety margins of the ammonia synthesis unit at each moment based on the wind and solar power generation power prediction curve and the process parameter change trends; determine the coupling relationship between the process parameters based on historical operating data and establish a safety constraint model for the process parameters; By combining the upper and lower limit constraints of the process parameters with the process parameter safety constraint model, the load adjustment space of the synthetic ammonia plant under different wind and solar power output conditions is obtained, including: According to the predicted value of the wind-solar power generation prediction curve at the current moment, the upper and lower limits of the raw gas pressure, the upper and lower limits of the circulating gas pressure, and the upper and lower limits of the synthesis tower temperature are determined to obtain the upper and lower limit constraints of the process parameters and the operating condition safety boundary; Extracting steady-state operating data of the raw gas pressure and the circulating gas pressure from historical operating data, calculating the linear correlation coefficient between the raw gas pressure and the circulating gas pressure, extracting dynamic operating data, determining the response time difference and response amplitude ratio between the synthesis tower temperature and the raw gas pressure and the circulating gas pressure, and inputting the linear correlation coefficient, response time difference, and response amplitude ratio into a safety constraint evaluation function to generate a process parameter safety constraint model; The upper and lower limit constraints of the process parameters and the operating condition safety boundaries are substituted into the process parameter safety constraint model, and the steady-state constraint intervals of the raw gas pressure and the circulating gas pressure, and the dynamic response constraint interval of the synthesis tower temperature are calculated respectively. The intersection of the steady-state constraint interval and the dynamic response constraint interval is taken to obtain the load regulation space of the synthetic ammonia unit under different wind and solar power output conditions.

4. The method according to claim 1, wherein Based on the load regulation space, the fluctuation pattern of historical process data is used as an optimization constraint in combination with the load optimization algorithm to generate an initial solution set for load distribution in each time period and calculate the fitness value. The pheromone release amount and volatility coefficient are dynamically adjusted according to the fitness value. When the change value of the optimal solution in continuous iteration is less than a preset convergence threshold, the time-based optimal load control instructions for the synthetic ammonia unit are output, including: Extract the continuous load change sequence in the historical process data, calculate the distribution interval of the absolute value of the load change and the absolute value of the load change rate at two adjacent time points, set the upper limit of the distribution interval as the upper limit of the load change amount and the upper limit of the load change rate, and obtain the load optimization constraint boundary; The load optimization algorithm is constructed based on the ant colony algorithm. Load solutions that meet the ant colony size requirements are randomly generated in the load regulation space. The load change amount and load change rate at adjacent time points in each set of load solutions are judged to be within the load optimization constraint boundary. The load solutions that meet the judgment conditions are retained as the initial load solution set. The load target deviation and load fluctuation amplitude of each load solution are calculated, and corresponding weights are assigned to the load target deviation and the load fluctuation amplitude respectively. The weighted sum result is used as the fitness value; Determine the pheromone baseline release amount according to the number of iterations completed so far, multiply the ratio of the optimal fitness value of the current iteration to the average fitness value by the pheromone concentration attenuation coefficient to obtain a pheromone concentration adjustment factor, adjust the baseline release amount based on the pheromone concentration adjustment factor to obtain an actual pheromone release amount, and update the actual pheromone release amount to the pheromone concentration matrix; A new round of load solution sets is generated according to the numerical distribution of the pheromone concentration matrix, the fitness value is repeatedly calculated and the pheromone concentration matrix is ​​updated. When the continuous iteration reaches a predetermined number of times and the change in the optimal fitness value is less than a preset convergence threshold, the corresponding load solution is output as the time-sharing optimal load control instruction for the synthetic ammonia plant.

5. The method according to claim 1, wherein Converting the optimal load control instruction into a control parameter, decomposing the control parameter into a process parameter target value, and implementing step-by-step adjustment through a cascade PID controller to achieve intelligent control of the synthetic ammonia production load includes: converting the optimal load control instruction into ammonia synthesis process control parameters, obtaining time series characteristics of the process parameters based on the response relationship between the parameters in historical operation data of the process parameters, and decomposing the ammonia synthesis process control parameters into process parameter target values ​​based on the time series characteristics; Using the process parameter target value as the given value of the main loop PID controller, calculating the deviation sequence between the actual output value of the main loop PID controller and the process parameter target value, integrating the deviation sequence with time weighted to obtain an integral time absolute error index, using the integral time absolute error index to calculate the adjustment direction and adjustment step size of the main loop PID controller parameters, updating the main loop proportional coefficient, integral coefficient and differential coefficient, and setting the updated main loop PID controller output value as the given value of the slave loop PID controller; The actual adjustment value of the process parameters is obtained by step-by-step adjustment of the cascade PID controller and sent to the corresponding actuator to realize intelligent load control of the synthetic ammonia unit.

6. The method according to claim 5, characterized in that The actual adjustment value of the process parameters is obtained by step-by-step adjustment of the cascade PID controller and sent to the corresponding actuator to achieve intelligent load control of the synthetic ammonia unit, including: Extracting dynamic response data between process parameters, statistically analyzing the fluctuation transmission relationship and response timing characteristics between process parameters, generating a coupling coefficient matrix and timing adjustment coefficients for the process parameters, superimposing the output value of the slave loop PID controller with the compensation values ​​of other loops calculated based on the coupling coefficient matrix to obtain an initial control variable, calculating the execution timing delay of each loop based on the timing adjustment coefficient, and performing hierarchical adjustment on the initial control variable according to the execution timing delay; collecting response characteristic curves of the actuator under different input signals, extracting gain characteristics and delay characteristics from the response characteristic curves, calculating a feedforward compensation gain based on the gain characteristics, determining a feedforward differential time based on the delay characteristics, applying the feedforward compensation gain and the feedforward differential time to an initial control variable after step-by-step adjustment, and obtaining a compensated execution instruction; The compensated execution instruction is issued to the actuator and the actual output value of the actuator is collected. The execution deviation sequence between the compensated execution instruction and the actual output value is calculated. A dynamic correction coefficient is determined according to the change trend of the execution deviation sequence. The execution instruction corrected based on the dynamic correction coefficient is reissued to the actuator to realize intelligent load control of the synthetic ammonia plant.

7. An intelligent control system for synthetic ammonia production load based on wind and solar power forecasting, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect data from wind farms and photovoltaic power plants as meteorological sample sets, and data from synthetic ammonia plants as process sample sets; The second unit is configured to process the meteorological sample set and the process sample set respectively using a dual-stream variational autoencoder, extract the temporal features of the meteorological data and the process data through an encoding network, generate latent features in combination with a residual connection mechanism, and generate a wind and solar power generation power forecast curve and a change trend of process parameters of the synthetic ammonia unit for a future period based on the latent features; The third unit is configured to calculate the upper and lower limit constraints of the process parameters and the operating safety boundary of the synthetic ammonia unit at each moment according to the wind and solar power generation power prediction curve and the process parameter change trend; Determine the coupling relationship between process parameters based on historical operating data and establish a safety constraint model for process parameters; combine the upper and lower limit constraints of the process parameters with the process parameter safety constraint model to obtain the load regulation space of the synthetic ammonia plant under different wind and solar power output conditions; A fourth unit is configured to generate an initial solution set for load distribution in each time period based on the load regulation space and a load optimization algorithm, using the fluctuation pattern of historical process data as an optimization constraint, calculate a fitness value, dynamically adjust the pheromone release amount and volatility coefficient according to the fitness value, and output a time-based optimal load control instruction for the ammonia synthesis unit when the change value of the optimal solution in continuous iterations is less than a preset convergence threshold; The fifth unit is used to convert the optimal load control instruction into a control parameter, decompose the control parameter into a process parameter target value, and realize intelligent control of the synthetic ammonia production load by step-by-step adjustment through a cascade PID controller.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

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

Images