Intelligent coal blending method based on prediction model

Abstract

The invention relates to an intelligent coal blending method based on a prediction model, and the method comprises the following steps: step 1, data collection and preprocessing, step 2, model training: a model adopts an Encoder-Decoder structure, an Encoder model is a state estimation model, and a Decoder model is a system dynamic model, and step 3, export finished coal quality control. According to the technical scheme, the quality stability of finished coal is obviously improved, it is ensured that each batch can accurately meet the strict requirements of customers, dependence on manual operation is greatly reduced, and therefore the labor cost is obviously reduced. More importantly, the intelligent coal blending system can accurately regulate and control the blending proportion of the raw material coal, and the situation that finished coal with the quality higher than the actual required quality level has to be produced in order to guarantee that the quality reaches the standard in the past is avoided, so that the production cost is further optimized while the quality is guaranteed, and double improvement of economical efficiency and quality is achieved.

Description

Technical Field

[0001] The present invention relates to an intelligent coal blending method, in particular to an intelligent coal blending method based on a prediction model, and belongs to the technical field of intelligent coal blending. Background Art

[0002] Coal blending is a clean and efficient coal utilization technology that mixes raw coals of different qualities in a certain blending ratio to obtain finished coal that meets the needs. The traditional coal blending method is for on-site operators to operate the coal feeder gate under the product bin to manually estimate and set the ratio of raw coal of different qualities. Due to the large fluctuations in the quality of raw coal in the product bin, the lag in the feedback of online ash detection results, and the time-varying and nonlinear transportation adaptability of the system, the traditional coal blending method cannot adjust the coal blending ratio in real time according to the changes in the quality of the finished coal, resulting in unstable quality of the finished coal and failure to meet customer needs. Therefore, a new solution is urgently needed to solve this technical problem. Summary of the invention

[0003] The present invention aims at the technical problems existing in the prior art and provides an intelligent coal blending method based on a prediction model. This technical solution improves the stability of commercial coal quality and avoids the waste of high-priced clean coal on the basis of meeting the needs of product users.

[0004] In order to achieve the above object, the technical solution of the present invention is as follows: an intelligent coal blending method based on a prediction model, the method comprising the following steps:

[0005] Step 1: Data collection and preprocessing,

[0006] Step 2: Model training. The model adopts the Encoder-Decoder structure, where the Encoder model is the state estimation model and the Decoder model is the system dynamic model.

[0007] Step 3: Quality control of finished coal for export.

[0008] Among them, step 1: data collection and preprocessing, as follows:

[0009] Step 1-1: Collect historical data of the intelligent coal blending process. The data set consists of several complete coal blending orders. The quality data q of the raw coal bunker is recorded in each order. i,j and production data that varies over time, where each time sample in the series includes the blending ratio and the quality of finished coal for export where i∈[1,N order ] represents the order number, j∈[1,N] represents the raw coal bunker number, t represents the time, q i,j represents the quality of the jth raw coal bunker in order i, represents the blending ratio of the jth raw coal bunker at time t in order i, represents the quality of the exported finished coal at time t in order i;

[0010] Step 1-2: Data preprocessing: center average filtering is performed on the quality data of finished coal exported to eliminate the measurement noise in the quality data of finished coal. The center average filtering method is:

[0011]

[0012] Where 2K+1 is the filter width, K is the preset value, k∈[tK,t+K] represents the quality of the exported finished coal at time k in order i, It represents the quality of the exported finished coal after filtering at time t in order i.

[0013] Among them, step 2: model training is as follows:

[0014] Step 2-1: The state estimation model predicts the current state of the system based on the system's recent input and output history sequences. Its form is as follows:

[0015]

[0016] in i∈[1,N order ] is the system status at time t in order i, i∈[1,N order ], j∈[1,N] represents the mixing ratio sequence data of the jth raw coal bunker from time t-T1 to time t-1 in order i, q i,j ,i∈[1,N order ],j∈[1,N] represents the quality of the jth raw coal bunker in order i, represents the quality sequence data of finished coal from time t-T1+1 to time t in order i, T1 is the first preset time length, θ is the parameter of the state estimation model,

[0017] Step 2-2: System dynamic model. The system dynamic model includes two sub-models: state transition model and output model. The state transition model describes the state transition characteristics of the system in the following form:

[0018]

[0019] in, is the system status at time t in order i, is the blending ratio of the jth raw coal bunker at time t in order i, q i,j is the mass of the jth raw coal bunker in order i, φ is the parameter of the state transition model,

[0020] Among them, the output model maps the system state to the controlled output value, and its form is as follows:

[0021]

[0022] in is the system status at time t in order i, is the system output at time t in order i, i.e. the predicted quality of finished coal, are the parameters of the output model.

[0023] The state estimation model is an encoder network from a sequence to a state vector, which is implemented by a network structure such as a transformer or LSTM. The system dynamic model is a decoder network that describes the dynamic characteristics of the system, which can be implemented using an LSTM network. When implementing, it is necessary to ensure that the dimensions of the state vectors in the encoder network and the decoder network are equal. The output model consists of a simple two-layer linear network, and a nonlinear activation function is used between the two linear layers to increase the nonlinearity of the model.

[0024] Among them, step 3: quality control of finished coal for export, as follows:

[0025] Based on the trained state estimation model and system dynamic model, the optimal blending ratio of each raw coal bunker is determined to achieve accurate control of the quality of exported finished coal. The blending ratio of each raw coal bunker meets certain constraints:

[0026] Constraints (1):

[0027] Constraint (2): in and They represent the minimum and maximum blending ratios of the j-th raw coal bunker in order i respectively.

[0028] Step 3 also includes the following constraints:

[0029] Constraint (3): When When the quality requirement of the finished coal exported by order i is less than that of the finished coal, the minimum blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, when When the quality requirement of the finished coal exported by order i is greater than that of the finished coal, the highest blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, Get the quality of the exported finished coal at time t After that, the following steps are performed to determine the optimal blending ratio, thereby controlling the quality of the exported finished coal to meet customer requirements: (1) Construct a historical production sequence. The historical production sequence in production order i includes the blending ratios of each raw coal bin from t-T1 to the current time t-1. i∈[1,N order ], j∈[1,N], the quality sequence of finished coal from time t-T1+1 to time t And the quality data of each raw coal bunker, q i,j ,i∈[1,N order ],j∈[1,N], where T1 is the first preset duration,

[0030] (2) Based on the constructed historical production sequence, the state estimation model is used to predict the system state at the current moment.

[0031] (3) Generate several blending strategies for raw coal bunkers. First, determine the minimum blending ratio for each bunker. and the highest blending ratio Next, for the first raw coal bunker to the N-1 raw coal bunker, in the interval i∈[1,N order ],j∈[1,N-1], uniformly sample to obtain m values; then, permutate and combine the mixing ratios of these N-1 coal bins to obtain m N-1 There are two blending strategies, among which the blending ratio of the Nth raw coal bunker can be determined by Get; finally, will not be satisfied The d strategies of the condition are deleted.

[0032] (4) Based on m N-1 -d doping strategies, generate m N-1 -d prediction output sequences, for each doping strategy, based on the current system status obtained in step 2 And the quality of each raw coal bunker i,j ,i∈[1,N order ],j∈[1,N], the system dynamic model and output model can be used to iteratively obtain a prediction output sequence, that is, from the current time t to the future time t+T2, the current blending ratio strategy is always executed to obtain the finished coal quality estimation sequence.

[0033] (5) First, from m N-1 -d finished coal quality estimation sequences are selected to find the finished coal quality setting value Q set The sequence that allows fluctuations within the range, that is The number of sequences that meet the conditions is recorded as ns; then, the difference between each sequence that meets the conditions and the set value of the finished coal quality is calculated. The sum of the absolute deviations of Secondly, calculate the corresponding product cost when executing the blending ratio of each sequence selected where p i,j ,i∈[1,N order ], j∈[1,N] is the price of the jth raw coal bunker in order i; then, calculate the overall target value Among them, α is the weight coefficient, which is used to balance the control effect and product cost; finally, the corresponding blending ratio with the smallest overall target value is selected as the decision result at the current time t.

[0034] An electronic device comprises: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the intelligent coal blending method based on a prediction model.

[0035] A computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the intelligent coal blending method based on a prediction model.

[0036] Compared with the prior art, the present invention has the following advantages: 1. The present invention proposes an intelligent coal blending method based on a prediction model, which determines the optimal blending ratio at the current moment based on the quality of each raw coal bunker, the blending ratio sequence data of each raw coal bunker and the quality sequence data of the finished coal, effectively overcoming the problems of large fluctuations in the quality of raw coal, time-varying and nonlinear operation of the system; 2. The invention significantly improves the quality stability of finished coal, ensures that each batch can accurately meet the strict requirements of customers, and greatly reduces the dependence on manual operation, thereby significantly reducing labor costs. It is expected that labor costs will be reduced by 20%-40%. In addition, the system can accurately control the blending ratio of raw coal, avoiding the production of finished coal higher than actual demand in order to ensure quality standards in the past, reducing raw material costs by about 5%-15% and operating costs by about 5%. In terms of production efficiency, intelligent scheduling and real-time monitoring can shorten the production cycle by 10%-30%. These improvements not only ensure quality compliance, but also further optimize production costs, achieve a dual improvement in economy and quality, and bring significant economic benefits and market competitive advantages to enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the overall process of the present invention,

[0038] Figure 2 It is a schematic diagram of the overall control process.

[0039] Figure 3 This is the actual control process diagram. DETAILED DESCRIPTION

[0040] In order to deepen the understanding of the present invention, the present embodiment is described in detail below with reference to the accompanying drawings.

[0041] Example 1: See Figure 1-Figure 3 , an intelligent coal blending method based on a prediction model, the method comprising the following steps:

[0042] Step 1: Data collection and preprocessing,

[0043] Step 2: Model training. The model adopts the Encoder-Decoder structure, where the Encoder model is the state estimation model and the Decoder model is the system dynamic model.

[0044] Step 3: Quality control of finished coal for export.

[0045] Among them, step 1: data collection and preprocessing, as follows:

[0046] Step 1-1: Collect historical data of the intelligent coal blending process. The data set consists of several complete coal blending orders. The quality data q of the raw coal bunker is recorded in each order. i,j and production data that varies over time, where each time sample in the series includes the blending ratio and the quality of finished coal for export Where i∈[1,Norder] represents the order number, j∈[1,N] represents the raw coal bunker number, t represents the time, q i,j represents the quality of the jth raw coal bunker in order i, represents the blending ratio of the jth raw coal bunker at time t in order i, represents the quality of the exported finished coal at time t in order i;

[0047] Step 1-2: Data preprocessing: center average filtering is performed on the quality data of finished coal exported to eliminate the measurement noise in the quality data of finished coal. The center average filtering method is:

[0048]

[0049] Where 2K+1 is the filter width, K is the preset value, k∈[tK,t+K] represents the quality of the exported finished coal at time k in order i, It represents the quality of the exported finished coal after filtering at time t in order i.

[0050] Among them, step 2: model training is as follows:

[0051] Step 2-1: The state estimation model predicts the current state of the system based on the system's recent input and output history sequences. Its form is as follows:

[0052]

[0053] in i∈[1,N order ] is the system status at time t in order i, i∈[1,N order ], j∈[1,N] represents the mixing ratio sequence data of the jth raw coal bunker from time t-T1 to time t-1 in order i, q i,j ,i∈[1,N order ],j∈[1,N] represents the quality of the jth raw coal bunker in order i, represents the quality sequence data of finished coal from time t-T1+1 to time t in order i, T1 is the first preset time length, θ is the parameter of the state estimation model,

[0054] Step 2-2: System dynamic model. The system dynamic model includes two sub-models: state transition model and output model. The state transition model describes the state transition characteristics of the system in the following form:

[0055]

[0056] in, is the system status at time t in order i, is the blending ratio of the jth raw coal bunker at time t in order i, q i,j is the mass of the jth raw coal bunker in order i, φ is the parameter of the state transition model,

[0057] Among them, the output model maps the system state to the controlled output value, and its form is as follows:

[0058]

[0059] in is the system status at time t in order i, is the system output at time t in order i, i.e. the predicted quality of finished coal, are the parameters of the output model.

[0060] The state estimation model is an encoder network from a sequence to a state vector, which is implemented by a network structure such as a transformer or LSTM. The system dynamic model is a decoder network that describes the dynamic characteristics of the system, which can be implemented using an LSTM network. When implementing, it is necessary to ensure that the dimensions of the state vectors in the encoder network and the decoder network are equal. The output model consists of a simple two-layer linear network, and a nonlinear activation function is used between the two linear layers to increase the nonlinearity of the model.

[0061] Based on the preprocessed data, a training data set is constructed. Each training sample in the training set contains three parts: the input of the state estimation model, the partial input of the system dynamic model, and the expected output of the output model. The input of the state estimation model includes i∈[1,N order ],j∈[1,N], where is the unfiltered export finished coal quality data, and the output of the state estimation model is The input of the system dynamics model is i∈[1,N order ],j∈[1,N], where Provided by the state estimation model, Provided by the dataset, the output of the system dynamics model is The input of the output model is The output of the output model is Where T1 and T2 are the first preset duration and the second preset duration respectively. Output based on the output model The expected output of the output model in the dataset The loss is calculated to update the parameters of the state estimation model, system dynamics model, and output model.

[0062] Among them, step 3: quality control of finished coal for export, as follows:

[0063] Based on the trained state estimation model and system dynamic model, the optimal blending ratio of each raw coal bunker is determined to achieve accurate control of the quality of exported finished coal. The blending ratio of each raw coal bunker meets certain constraints:

[0064] Constraints (1):

[0065] Constraint (2): in and They represent the minimum and maximum blending ratios of the j-th raw coal bunker in order i respectively.

[0066] Step 3 also includes the following constraints:

[0067] Constraint (3): When When the quality requirement of the finished coal exported by order i is less than that of the finished coal, the minimum blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, when When the quality requirement of the finished coal exported by order i is greater than that of the finished coal, the highest blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, Get the quality of the exported finished coal at time t Finally, the following steps are performed to determine the optimal blending ratio, thereby controlling the quality of the exported finished coal to meet customer requirements.

[0068] (1) Construct the historical production sequence. The historical production sequence in production order i includes the blending ratios of each raw coal bunker from t-T1 to the current time t-1. i∈[1,N order ], j∈[1,N], the quality sequence of finished coal from time t-T1+1 to time t And the quality data of each raw coal bunker, q i,j ,i∈[1,N order ],j∈[1,N], where T1 is the first preset duration,

[0069] (2) Based on the constructed historical production sequence, the state estimation model is used to predict the system state at the current moment.

[0070] (3) Generate several blending strategies for raw coal bunkers. First, determine the minimum blending ratio for each bunker. and the highest blending ratio Next, for the first raw coal bunker to the N-1 raw coal bunker, in the interval i∈[1,N order ],j∈[1,N-1], uniformly sample to obtain m values; then, permutate and combine the mixing ratios of these N-1 coal bins to obtain m N-1 There are two blending strategies, among which the blending ratio of the Nth raw coal bunker can be determined by Get; finally, will not be satisfied The d strategies of the condition are deleted.

[0071] (4) Based on m N-1 -d doping strategies, generate m N-1 -d prediction output sequences, for each doping strategy, based on the current system status obtained in step 2 And the quality of each raw coal bunker i,j ,i∈[1,N order],j∈[1,N], the system dynamic model and output model can be used to iteratively obtain a prediction output sequence, that is, from the current time t to the future time t+T2, the current blending ratio strategy is always executed to obtain the finished coal quality estimation sequence.

[0072] (5) First, from m N-1 -d finished coal quality estimation sequences are selected to find the finished coal quality setting value Q set The sequence that allows fluctuations within the range, that is The number of sequences that meet the conditions is recorded as ns; then, the difference between each sequence that meets the conditions and the set value of the finished coal quality is calculated. The sum of the absolute deviations of Secondly, calculate the corresponding product cost when executing the blending ratio of each sequence selected where p i,j ,i∈[1,N order ], j∈[1,N] is the price of the jth raw coal bunker in order i; then, calculate the overall target value Among them, α is the weight coefficient, which is used to balance the control effect and product cost; finally, the corresponding blending ratio with the smallest overall target value is selected as the decision result at the current time t.

[0073] It should be noted that the above embodiments are not intended to limit the protection scope of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. An intelligent coal blending method based on a prediction model, characterized in that: The method comprises the following steps: Step 1: Data collection and preprocessing, Step 2: Model training. The model adopts the Encoder-Decoder structure, where the Encoder model is the state estimation model and the Decoder model is the system dynamic model. Step 3: Quality control of finished coal for export.

2. The intelligent coal blending method based on the prediction model according to claim 1 is characterized in that: Step 1: Data collection and preprocessing, as follows: Step 1-1: Collect historical data of the intelligent coal blending process. The data set consists of several complete coal blending orders. The quality data q of the raw coal bunker is recorded in each order. i,j and production data that varies over time, where each time sample in the series includes the blending ratio and the quality of finished coal for export where i∈[1,N order ] represents the order number, j∈[1,N] represents the raw coal bunker number, t represents the time, q i,j represents the quality of the jth raw coal bunker in order i, represents the blending ratio of the jth raw coal bunker at time t in order i, represents the quality of the exported finished coal at time t in order i; Step 1-2: Data preprocessing: center average filtering is performed on the quality data of finished coal exported to eliminate the measurement noise in the quality data of finished coal. The center average filtering method is: Where 2K+1 is the filter width, K is the preset value, k∈[tK,t+K] represents the quality of the exported finished coal at time k in order i, It represents the quality of the exported finished coal after filtering at time t in order i.

3. The intelligent coal blending method based on the prediction model according to claim 2 is characterized in that: Step 2: Model training is as follows: Step 2-1: The state estimation model predicts the current state of the system based on the system's recent input and output history sequences. Its form is as follows: in i∈[1,N order ] is the system status at time t in order i, i∈[1,N order ], j∈[1,N] represents the mixing ratio sequence data of the jth raw coal bunker from time t-T1 to time t-1 in order i, q i,j ,i∈[1,N order ],j∈[1,N] represents the quality of the jth raw coal bunker in order i, represents the quality sequence data of finished coal from time t-T1+1 to time t in order i, T1 is the first preset time length, θ is the parameter of the state estimation model, Step 2-2: System dynamics model. The system dynamics model includes two sub-models: state transition model and output model. Among them, the state transition model describes the state transition characteristics of the system, and its form is as follows: in, is the system status at time t in order i, is the blending ratio of the jth raw coal bunker at time t in order i, q i,j is the mass of the jth raw coal bunker in order i, φ is the parameter of the state transition model, Among them, the output model maps the system state to the controlled output value, and its form is as follows: in is the system status at time t in order i, is the system output at time t in order i, i.e. the predicted quality of finished coal, are the parameters of the output model.

4. The intelligent coal blending method based on the prediction model according to claim 3 is characterized in that: Step 3: Quality control of finished coal for export, as follows: Based on the trained state estimation model and system dynamic model, the optimal blending ratio of each raw coal bunker at the current time t is determined, so as to achieve accurate control of the quality of exported finished coal. The blending ratio of each raw coal bunker meets certain constraints: Constraints (1): Constraint (2): in and They represent the minimum and maximum blending ratio of the j-th raw coal bunker in order i respectively.

5. The intelligent coal blending method based on the prediction model according to claim 4 is characterized in that: Step 3 also includes the following constraints: Constraint (3): When When the quality requirement of the finished coal exported by order i is less than that of the finished coal, the minimum blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, when When the quality requirement of the finished coal exported by order i is greater than that of the finished coal, the highest blending ratio of the best quality raw coal is the current blending ratio of the coal of this category, that is, Get the quality of the exported finished coal at time t After that, the following steps are performed to determine the optimal blending ratio, thereby controlling the quality of the exported finished coal to meet customer requirements: (1) Construct a historical production sequence. The historical production sequence in production order i includes the blending ratio sequence of each raw coal bin from t-T1 to t-1. i∈[1,N order ], j∈[1,N], the quality sequence of finished coal from time t-T1+1 to time t And the quality data of each raw coal bunker i,j ,i∈[1,N order ],j∈[1,N], where T1 is the first preset duration, (2) Based on the constructed historical production sequence, the state estimation model is used to predict the system state at the current moment. (3) Generate several blending strategies for raw coal bunkers. First, determine the minimum blending ratio for each bunker. and the highest blending ratio Next, for the first raw coal bunker to the N-1 raw coal bunker, in the interval i∈[1,N order ],j∈[1,N-1], uniformly sample to obtain m values; then, permutate and combine the mixing ratios of these N-1 coal bins to obtain m N-1 There are two blending strategies, among which the blending ratio of the Nth raw coal bunker can be determined by Get; finally, will not be satisfied The d strategies of the condition are deleted. (4) Based on m N-1 -d doping strategies, generate m N-1 -d prediction output sequences, for each doping strategy, based on the current system status obtained in step 2 And the quality of each raw coal bunker i,j ,i∈[1,N order ],j∈[1,N], the system dynamic model and output model can be used to iteratively obtain a prediction output sequence, that is, from the current time t to the future time t+T2, the current blending ratio strategy is always executed to obtain the finished coal quality estimation sequence. (5) First, from m N-1 -d finished coal quality estimation sequences are selected to find the finished coal quality setting value Q set The sequence that allows fluctuations within the range, that is The number of sequences that meet the conditions is recorded as ns; then, the difference between each sequence that meets the conditions and the set value of the finished coal quality is calculated. The sum of the absolute deviations of Secondly, calculate the corresponding product cost when executing the blending ratio of each sequence selected where p i,j ,i∈[1,N order ], j∈[1,N] is the price of the jth raw coal bunker in order i; then, calculate the overall target value Among them, α is the weight coefficient, which is used to balance the control effect and product cost; finally, the corresponding blending ratio with the smallest overall target value is selected as the decision result at the current time t.

6. An electronic device, characterized in that: include: a processor, a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the intelligent coal blending method based on the prediction model as described in any one of claims 1 to 5.

7. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, an intelligent coal blending method based on a prediction model as described in any one of claims 1 to 5 is implemented.

Images